WO2011009011A1 - An integrated framework for vehicle operator assistance based on a trajectory prediction and threat assessment - Google Patents
An integrated framework for vehicle operator assistance based on a trajectory prediction and threat assessment Download PDFInfo
- Publication number
- WO2011009011A1 WO2011009011A1 PCT/US2010/042203 US2010042203W WO2011009011A1 WO 2011009011 A1 WO2011009011 A1 WO 2011009011A1 US 2010042203 W US2010042203 W US 2010042203W WO 2011009011 A1 WO2011009011 A1 WO 2011009011A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- threat
- vehicle
- control
- optimal
- operator
- Prior art date
Links
Classifications
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B9/00—Simulators for teaching or training purposes
- G09B9/02—Simulators for teaching or training purposes for teaching control of vehicles or other craft
- G09B9/04—Simulators for teaching or training purposes for teaching control of vehicles or other craft for teaching control of land vehicles
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/10—Geometric CAD
- G06F30/15—Vehicle, aircraft or watercraft design
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F30/00—Computer-aided design [CAD]
- G06F30/20—Design optimisation, verification or simulation
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B19/00—Teaching not covered by other main groups of this subclass
- G09B19/16—Control of vehicles or other craft
- G09B19/167—Control of land vehicles
-
- G—PHYSICS
- G09—EDUCATION; CRYPTOGRAPHY; DISPLAY; ADVERTISING; SEALS
- G09B—EDUCATIONAL OR DEMONSTRATION APPLIANCES; APPLIANCES FOR TEACHING, OR COMMUNICATING WITH, THE BLIND, DEAF OR MUTE; MODELS; PLANETARIA; GLOBES; MAPS; DIAGRAMS
- G09B9/00—Simulators for teaching or training purposes
- G09B9/02—Simulators for teaching or training purposes for teaching control of vehicles or other craft
- G09B9/04—Simulators for teaching or training purposes for teaching control of vehicles or other craft for teaching control of land vehicles
- G09B9/048—Simulators for teaching or training purposes for teaching control of vehicles or other craft for teaching control of land vehicles a model being viewed and manoeuvred from a remote point
Definitions
- inventions described herein relate to a novel, corridor-based framework that performs threat assessment and provides varying degrees of mutually consistent automated operator assistance as a threat response in human-machine systems, such as locally or remotely-operated passenger vehicles, transport vehicles, agricultural machinery, fork lift trucks, aerial vehicles, robots, or surgical tools.
- This framework explicitly considers human and machine dynamics without presuming operator intentions or limiting the
- avoidance maneuver and its associated threat assessment
- It provides a unified framework that allows for various modes and levels of mutually consistent operator assistance, from operator warning to stability control to passive intervention, to active semi-autonomous control, and finally, to autonomous machine operation.
- Automotive active safety systems are concerned with preventing accidents through the introduction of various computer-controlled actuation methods to inform, improve, or override a human operator's steering and/or braking
- Active safety systems currently in existence include yaw stability control, roll stability control, traction control, and antilock braking, among others. While these systems reduce accident frequency, their path-based and largely reactive nature limits their ability to: 1) accurately assess the threat posed by a given scenario and 2 ) adequately determine when and how to intervene to assist the driver. This dependence on a specific trajectory (amidst a myriad of options available to the operator) reduces the accuracy and significance of the threat assessment and leads to controllers that selectively replace (rather than assist) the driver in order to follow an automation-designated path.
- inventions described and claimed herein relate, primarily, to the unified nature of the framework, and its flexibility in being able to accurately assess threat, and then participate in one or more of a wide variety of mutually consistent operator assistance modes, of varying levels of operator autonomy, from complete autonomy, to no autonomy.
- Inventions described and claimed in the Threat assessment application relate primarily to threat assessment aspects of this novel framework. Both have similar disclosures.
- sensing systems such as radar, LIDAR, cameras, inertial measurement units and GPS localization systems are used to detect, classify, and track the position of objects and the drivable road surface in the host vehicle's vicinity as well as measure vehicle states. Once these potential hazards have been identified, localized and their motion has been
- a threat metric is used to quantify the threat they pose to the host vehicle, together with the threat of
- threat assessment technology is designed to then trigger and/or implement countermeasures to reduce the threat. These countermeasures can be passive or active. The effectiveness of threat assessment metrics depends on their ability to correctly identify hazards and accurately assess the threat that potential hazards pose to the host vehicle.
- Time-based threat measures project time to collision (TTC) based on current speeds, positions,
- Distance-based metrics are generally calculated using
- acceleration-based metrics assess the threat of a given maneuver based on the minimum (and often assumed
- At least one known method of assessing threat relates to a vehicle that is intended to navigate along a path.
- the path may be predetermined, or calculated, based on data, such as information about obstacles and a path followed by a track, such as a roadway in the case of a road vehicle, such as an automobile.
- the path is a curve of simple
- the system determines that danger has arisen, and the system generates a threat signal.
- the threat of actual danger is potentially low, because vehicles driven by a human operator typically operate within a field of safe travel, or a corridor, rather than along a relatively arbitrary line, such as the centerline of a roadway.
- This dependence on a specific trajectory reduces the accuracy and significance of the threat assessment and leads to controllers that selectively replace (rather than assist) the driver in order to follow an automation-designated path.
- a path is a simple geometric curve in two-dimensional space, along which a vehicle may travel.
- a path has a width of essentially zero.
- a trajectory is a physically-achievable and time-parameterized sequence of vehicle states (such as
- trajectory includes the velocity of a vehicle as an element. It may also be thought of as having time as a parameter of the path, which may then establish velocity at different locations.
- a corridor may be considered to be the space between two curves in two-dimensional space. Travel anywhere within the corridor is considered to be safe.
- a region is a concept that is defined in connection with inventions hereof, and it will be defined below.
- a Model Predictive Controller is an optimal control method typically used to generate an optimal set of control inputs (spanning through a future time horizon) required to track a desired path while minimizing a user-defined objective function.
- MPC Model Predictive Controller
- a human driver typically operates a vehicle within a safe range of vehicle states. For example, the driver
- a human operator operates within an tridimensional region of state space, rather than along a simple, zero, or nearly-zero width curve of a physical trajectory.
- Some known threat assessors exist as a separate system, not integrated with other systems of the device analysis and control apparatus. These systems may base the assessment of threat on a device state exceeding a relatively arbitrary threshold. Further, these threat assessors
- rudimentary threat assessment metrics based on, for example, the current deviation of the device from a predetermined optimal path. It would be desirable to be able to take
- a further object would be to be able to
- Still another object would be to be able to assess threat on a realistic corridor, rather than an unrealistic single path.
- Another object would be to assess threat with an apparatus that is integrated with other systems of system analysis and control, which are also used to control the device or system, rather than assessing threat on a device state exceeding an arbitrary threshold.
- Still another object would be to use the assessment of threat to generate an operator assistance signal to assist the operator in safely operating the device or replace the operator as necessary to ensure safe operation of the device.
- Fig. 1 is a block diagram illustrating basic
- Fig. 2 graphically shows an example of various potential intervention laws based on threat metric
- Fig. 3a showing the different levels of intervention
- Fig. 3b showing the relative locations of the host vehicle and the environment
- FIG. 4 shows, in flowchart form, basic steps of a method of threat assessment and semi-autonomous control, with possible considerations at each step;
- Figs. 5a, 5b and 5c show, graphically, a simulated test illustrating system response when driver fails to
- Figs. 6a, 6b and 6c show, graphically, a simulated test illustrating system response to an erroneous driver swerve, as shown schematically in Fig. 6a, with steering angle shown in Fig. 6b and control authority K shown in Fig. 6c, where K represents the proportion of control authority given to autonomous system, with the driver allowed the remaining (1-K);
- Figs. 7a, 7b and 7c show, graphically, a simulated test illustrating system response when driver fails to
- FIG. 8 shows, in flow chart form, steps of a method of an invention hereof for assessing threat, with additional considerations associated with some of the steps;
- FIG. 9a shows, schematically, an illustration of a simple path tracking control set up of the prior art
- Fig. 10 is a schematic representation in block diagram form illustrating how a representative system of the prior art that can lead to poor, mutually inconsistent
- Fig. 11 is a schematic representation in block diagram form showing an array of functions performed in a unified, mutually consistent manner by inventions hereof.
- inventions described in the above-referenced Semi- Autonomous applications, and the Comprehensive Provisional patent application relate, among other things, to a unified framework for performing threat assessment and semi-autonomous vehicle navigation and control while allowing for adaptable and configurable intervention laws and configurable control inputs .
- Inventions described in the Threat Assessment application relate to methods and apparati for identifying and quantifying threat being experienced by a system that can be modeled, such as a vehicle, such as a road vehicle, such as an automobile.
- a model of the device with a model of the environment and the current state of the device and the environment are used to iteratively generate a
- sequence of optimal device control inputs that, when applied to a model of the device, generate an optimal device
- This optimal trajectory and the sequence of device control inputs that generates it is used to generate a threat assessment metric.
- An appropriate type and level of operator assistance is generated based on this threat assessment.
- Operator assistance modes include warnings, decision support, operator feedback, vehicle stability
- each assistance mode is mutually consistent because they are generated using the same optimal trajectory.
- identifying and quantifying threat will be referred to as assessing threat, or sometimes,
- the methods and apparati for assessing threat are of the same type as can be used in the unified framework for semi-autonomous vehicle navigation and control, discussed in the applications mentioned above. Such methods and apparati can be used for assessing threat in any system that can be modeled.
- the inventions described herein do not rely on a subjective prediction of a path intended by a human operator or a specific path proposed by an automatic path planning algorithm. Instead, these inventions base threat assessment and operator assistance decisions on dynamic properties and known constraints inherent to the vehicle and the environment. These inventions warn of deviation from a physically-constrained and dynamically- feasible region in N space (that includes a two-dimensional corridor in physical space) rather than an arbitrarily- calculated path, which is often less meaningful since it often does not represent a true, or even achievable, much less optimal route. Further, by predictively simulating the vehicle dynamics over a time horizon, which may be finite or infinite, its threat assessment and intervention operations explicitly and pre-emptively consider the combined effects of vehicle dynamics, stability constraints, and terrain interaction on maneuver severity.
- a region is an N-dimensional analogue of a corridor.
- a region is an n-dimensional area defined in the state space of the device model over a time horizon. This region is bounded by corridor constraints (which apply to vehicle position states), together with other state
- constraints such as those imposed on vehicle states such as yaw angle, yaw rate, velocity, wheel sideslip angle, etc.).
- MPC planners of inventions hereof are not required to utilize a single
- planners described herein generate a course of motion that need not follow a predefined path but may instead generate its own optimal trajectory (and the control inputs necessary to achieve it) at successive time steps.
- Threat assessment inventions described herein may be used with any system that can be modeled, including vehicles, such as terrestrial, nautical, and aerial vehicles,
- the explanation is for illustrative purposes only, and that the inventions described herein can be used to identify and quantify (i.e., assess) threat in all systems that can be modeled.
- the systems may be physical, such as devices and chemical processes. They may also be non-physical, such as economic systems.
- the system model will include states (such as pitch angle for an aerial vehicle, solvent concentration for a chemical system, or asset prices for a financial system) for which a desired value or range of values exists and to which constraints (such as stall limits for an aerial vehicle, saturation for a chemical process, or price caps for a financial system) may apply.
- Sensory information would include data related to nearby vehicles, pedestrians, road edges, and other salient features to assess accident threat.
- navigation systems ideally operate only during instances of significant threat: it should give a driver full control of the vehicle in low threat situations but apply appropriate levels of computer-controlled actuator effort during high threat situations.
- An active navigation system can therefore be termed semi-autonomous, since it must allow for human- controlled, computer-controlled, and shared human/computer vehicle operation.
- Such a system should be as unobtrusive to the driver as possible (i.e. it should intervene only as much as is minimally required to avoid an impending accident).
- Fig. 1 shows, schematically, in block diagram form, a basic framework operation.
- a model Predictive Control (MPC) vehicle navigation block 110 starts with a model 112 of the environment, a model 114 of the device, in this case, a vehicle, and the vehicle's current state, including the position. It generates an optimal vehicle trajectory from the current position through a time horizon. The trajectory is optimal with respect to a predefined, configurable set of criteria. It also generates a corresponding optimal set of control input commands necessary to execute an optimal trajectory within the corridor and ensure that the vehicle 102 operates within safe driving limits (defined by a constraint-bounded region in the state space).
- the environment model can be based on a priori known information (e.g. from maps) and/or information gathered by real time sensors, such as on-vehicle sensors 104 (e.g.
- V2V sensors vehicle to vehicle
- information 106 related to the environmental and potential environmental hazards, such as position of road edges, lane boundaries, holes, slopes, static obstacles (e.g. trees, road-side signs), and dynamic obstacles (e.g. other vehicles, pedestrians).
- the vehicle model is user-defined and can be of varying complexity and fidelity.
- the real-time sensors may also be mounted in the environment and communicate with the control system on the vehicle.
- control inputs can be associated with one or multiple actuators, such as active steering, active braking, and others.
- the predicted vehicle trajectory and associated control inputs may be generated via constrained optimal control, which leverages efficient optimization methods and constraint-handling
- trajectory is used herein to mean a sequence of vehicle states, including its position, velocity, sideslip and yaw angles, etc.
- the predicted vehicle trajectory and predicted control inputs are analyzed by a threat assessor 108 to quantify the threat to the vehicle by computing a configurable metric, such as the maximum lateral acceleration, sideslip angle, or roll angle over the trajectory, the minimum proximity to obstacles, or other metrics.
- a configurable metric such as the maximum lateral acceleration, sideslip angle, or roll angle over the trajectory, the minimum proximity to obstacles, or other metrics.
- Threat may be more formally considered to be a hierarchical combination of obstacle avoidance, stability-critical states, inputs, etc, based on a model of the vehicle.
- the control authority exerted by the system can then be determined as a function of this generated threat: generally speaking, if the threat metric value is low, the control system intervention is low (i.e. the driver commands the vehicle with little or no computer-controlled intervention); if the threat metric value is high, the control system intervention is high.
- the form of the intervention law modulating this control system authority is configurable and can differ for different actuators (i.e. a vehicle with both active steering and braking can have
- intervention law can also be defined to adapt to driver performance based on an assessment of driver skill, and/or to include considerations for driver preference, environmental conditions, previous threat metric values, previous control inputs, and other factors.
- Fig. 2 shows, schematically, examples of various potential
- intervention laws showing, from top to bottom, linear, smooth and threshold-shaped intervention laws that depend only on predicted threat.
- the vertical axis represents the degree of control authority given to the active navigation and control system while the horizontal axis represents the predicted threat, with cause for intervention increasing from left to right.
- the control system begins to assume control authority to preempt an unsafe maneuver.
- the controller's authority phases out. In this manner, the system can be said to be both predictive and semi-autonomous .
- intervention law can be designed such that it assumes full authority by the time the predicted safe trajectory reaches the limit of any pre-defined critical vehicle states. This corresponds to a situation where only an optimal set of inputs would result in a safe vehicle trajectory.
- FIG. 3 shows schematically an obstacle avoidance scenario illustrating different stages of intervention for an inattentive driver. Initially, the host automobile 302 is at location 1. No obstacles are in view, and the optimal
- the predicted trajectory is a straight path.
- the predicted threat is at a low level, indicated by the vertical line designated 1, near to the left hand side, which represents low threat.
- the vehicle 302 advances along the roadway to the location 2, it comes nearer to a truck 304, whose velocity is either zero, or much less than that of the host vehicle 302.
- the sensors sense the proximity of the obstacle vehicle 304, and generate a threat metric that is larger, as at the vertical line designated 2, near the right hand limit of the threat scale.
- the optimal predicted trajectory assumes a curved shape, to avoid the obstacle 304. Simultaneously, the level of
- intervention K increases (as shown by each of the three different intervention laws) so that, in a semi-autonomous system, the controller would take more and more control, the nearer to the obstacle the host vehicle 302 comes.
- the controller may take different countermeasures, such as initiating a warning, priming brakes, seatbelts, or airbags, and/or engaging active systems, etc.
- Fig. 4 shows, schematically, in flow chart form, a basic flow of steps performed by a controller of an invention described in the Semi-Autonomous applications, above, with possible considerations at each step.
- An initial step 402 generates an optimal set of control inputs and corresponding vehicle trajectory by forward simulation. Considerations 402a for this step include, for example, (but are not limited to) the vehicle dynamics, current state of the vehicle and environment, terrain and environmental disturbances, available actuation, trajectory objectives, safety limits, and driver inputs.
- a next step is to assess 404 the predicted threat to the vehicle.
- General considerations 404a for this step include characteristics of the optimal path and associated control input, safety limits and driver inputs. The method of threat assessment is discussed below in more detail in connection with Fig. 8.
- a next step is to generate D06 control authority gains, with a major consideration D06a at this stage being the desired intervention characteristic.
- the next step D08 is to implement the scaled control for the current time.
- FIGs. 5a, 5b and 5c show, graphically, the results of a simulated test illustrating system response when a driver fails to navigate a curve in the road, shown by in Fig. 5a by a pair of lines.
- the trajectory that the driver would have followed without assistance is shown dashed. Note that it leaves the roadway. With assistance, it is shown solid black and remains within the roadway.
- K represents proportion of control authority given to the autonomous system, with the driver allowed the remaining (1-K).
- the middle graph, Fig. 5b shows the steer inputs, with the dashed line corresponding to the driver and the solid curve
- Fig. 5c shows the control authority given to the autonomous system, in this case, steering, with the degree varying with distance (x) along the horizontal scale.
- Figs. 6a, 6b and 6c show, graphically, the results of a simulated test illustrating the system response to an erroneous driver swerve.
- K represents proportion of control authority given to autonomous system, with the driver allowed the remaining (1-K).
- the same line types as above correspond to the driver without assistance (gray dashed) and with assistance (solid line).
- the safe roadway corridor is shown in Fig. 5a by a pair of light solid lines in the upper graph. Distance is shown along the horizontal scale. The assisted trajectory remains within the safe roadway.
- Figs. 7a, 7b and 7c show, graphically, a simulated test illustrating system response when a driver fails to anticipate/avoid an obstacle, similar to the scenario
- K represents the proportion of control authority given to autonomous system.
- the obstacle is simulated by a jog in the light line that represents one edge of the safe roadway. The only inputs used in this simulation are, again, steering of the driver and the autonomous system.
- Threat assessment techniques disclosed herein, and uses within a system as described herein provide improved modularity and adaptability when compared to previous methods and apparati. Aspects of this improved modularity are
- the underlying control framework can accommodate multiple actuation modes and vehicle models, allowing for ready application of the system to various vehicle types and actuator configurations.
- the system's intervention law is also readily adapted (i.e. it can change over time based on an assessment of driver skill, driver preference, environmental conditions, previous threat metric values, previous control inputs, and other factors). These adaptations can be performed either statically or dynamically.
- Gauging threat using metrics based on predicted vehicle state evolution within a region/corridor is novel. This includes threat assessment based on a trajectory that remains within that corridor, along with semi-autonomous control necessary to keep the vehicle within the safe
- corridor it is generally meant a portion of physical space, such as the width of a roadway, or a roadway and adjoining break down lane.
- region it is generally meant a region in N-space, in which two of these N dimensions may be the two spatial dimensions of the roadway's width and length, and other dimensions may be states and/or control inputs of the vehicle, such as sideslip angle, yaw angle, velocity, steering angle, etc. This leads to significant difference in performance of the two general approaches.
- the simple path-based approach of the prior art is suited only to warn or selectively replace a human
- the corridor/region approach described herein may warn, supplement, or act in conjunction with a human operator, as well as replace.
- FIG. 8 a representative process for assessing threat is described. The steps about to be
- Considerations 802a that are taken in conjunction with this step may include, but are not limited to: vehicle dynamics and constraints, trajectory objectives (controller objective function), terrain and environmental disturbances and environmental information, such as: obstacle locations, terrain properties (slope, roughness, friction coefficient, etc.) and other disturbances (such as wind) . It is through these considerations of vehicle dynamics and constraints and obstacle locations, that the corridor aspects of this method of threat assessment enters. Trajectory objectives more broadly describes the penalties applied to violating constraints and/or approaching unstable states and shapes what an optimal solution looks like within the
- the model predictive controller generates an optimal trajectory, which constitutes an optimal sequence of inputs and the corresponding set of optimal vehicle states over a time horizon. These outputs are generated by the MPC unit, based on analysis of the models of the vehicle and the
- model predictive control over a time horizon, as discussed in more detail below.
- threat assessment methods disclosed herein operate on an optimal (sometimes referred to herein and the literature as best case, or benchmark)
- a desired path gives a rather inflexible and in some cases, arbitrary goal state to which the driver will be forced to adhere or risk setting off warning indicators or controller intervention
- the optimal trajectory gives an indication of the threat posed to the vehicle and a backup plan in case the human operator doesn't keep the vehicle within the
- the optimal trajectory, of predicted states and predicted vehicle control inputs, over a time horizon, is next coalesced 804 to generate a scalar prediction metric.
- Considerations 804a that contribute to the step of generating a scalar metric may include but are not limited to: past, present and/or predicted states, inputs and objective function costs.
- Various norms discussed below may be used to combine trajectory components into this scalar.
- Past present and/or predicted operator inputs and performance may also, but need not be taken into account.
- the result of the coalescing step 804 is a scalar prediction metric.
- This scalar can be non-dimensionalized 806 by normalizing it against known or approximated physical limits and/or predetermined desired thresholds to obtain a threat assessment at the current time. Suitable candidates for such thresholds include, but are not limited to maximum sideslip angle before loss of control, maximum load transfer before wheel liftoff, maximum lateral acceleration before skidding, maximum longitudinal acceleration before skidding, maximum total acceleration before skidding, maximum steer angle before actuator saturation, maximum available
- Considerations 806a that may be taken into account include, but are not limited to known physical limits on vehicle dynamics (e.g. friction saturation limits, rollover thresholds, etc), desired intervention characteristics, operator performance and operator preference.
- the result of the non-dimensionalizing step 806 is a threat assessment, which can then be used for one or more of a variety of mutually consistent threat response functions, depending upon the system configuration.
- the threat assessment can be used to generate a warning, which may be perceptible by any human sense, including audible, visual, haptic and
- the threat assessment can also, or alternatively, be used to trigger assistance, which may be passive, or active, as discussed above, and, if active, to varying
- the threat assessment may even be used to modulate other system or vehicle characteristics.
- Examples of passive assistance in the context of vehicular control include, but are not limited to: resistance torques on steering wheel, traction control, anti-lock braking systems.
- Examples of active assistance include, but are not limited to adaptive cruise control, yaw stability assistance Electronic Stability Control (ESC) lane keeping assistance, obstacle avoidance. Modulation of other system characteristics may include but are not limited to: seatbelt pretensioning, brake priming, suspension modifications (active suspensions, suspension stiffness modifications, etc.).
- Examples of semi- autonomous (a form of active assistance) assistance include, but are not limited to partial control of the steering and braking commands sent to the vehicle.
- Examples of autonomous assistance include, but are not limited to, full control of one or more vehicle actuator such as wheel steering angles, braking torques, acceleration torques, etc.
- Model Predictive (MPC) (or receding horizon) Control is a family of finite-horizon optimal control schemes that iteratively minimizes a performance objective defined for a forward- simulated plant model subject to performance and input
- MPC uses a model of the plant to predict future vehicle state evolution and optimize a set of plant control inputs such that this prediction satisfies constraints and minimizes a user-defined objective function.
- Model predictive control has a number of significant properties that make it particularly well suited to threat assessment for use with, for instance, autonomous and semi- autonomous vehicle navigation problems. Its ability to
- MPC can be configured to according to inventions disclosed herein, to plan its own, optimal, path given a set of
- the path thus planned through the (pre-delineated) safe operating environment potentially offers a number of advantages over alternative trajectory planning methods; not only is it explicitly aware of vehicle dynamics, measured disturbances, and actuator limitations, but the constraint-satisfying trajectory plan it generates is feasible, since it is obtained from an already- generated set of control inputs. It is also optimal, with respect to some performance metric such as minimum lateral acceleration over a future time horizon, minimum wheel slip, etc. In the semi-autonomous framework described below, this optimal prediction can serve not only as a optimal trajectory plan, but also as an effective threat assessor.
- the first control element of this input sequence is implemented at the current time and the process is repeated at subsequent time steps. A state sequence spanning the same time period is also generated. No known MPC process uses elements of either of these sequences, after it has identified the single control input element for the current time. Only that single, current control input element is used.
- inventions disclosed herein use these subsequent elements of one or both the state and control sequences to assess threat to which the device is (or may in the future be) exposed.
- controller inputs also enables generation of a metric for analyzing the threat posed to the device by a given scenario.
- This metric is comparable to and in many situations more useful/accurate than existing metrics because it is based solely on known or approximated physical limits of the
- threshold threat values may trigger driver warnings at critical/desired threat thresholds.
- threat assessment may be used to determine when and how strongly to intervene. The latter application is a topic of the Semi-Autonomous applications. This disclosure focuses on the design and development of the threat assessment metric itself .
- a corridor-based trajectory-planning method may be used, based on constrained optimal control. When the objective function and constraints are defined as described below, the vehicle path calculated at each time step by the MPC
- controller is assumed to be the best case or safest path within a corridor through the environment. Some key metrics from this prediction may be used to assess the instantaneous threat posed to the vehicle.
- y k Cx, + D v v, with x, y, u, and v representing states, outputs, inputs, and disturbances of the system respectively, a quadratic
- the vector V allows for variable constraint softening over the prediction horizon, p, when ⁇ is included in the objective function.
- trajectory-tracking setup of a prior art controller and threat assessor, through a hazard-containing environment may then be illustrated by Fig. 9a.
- Fig. 9a shows a vehicle 902a that seeks to avoid an obstacle of another vehicle 904a and a pedestrian 906a.
- a path based controller and threat assessor would attempt to follow a desired single track path y des . This path may safely navigate the hazards, but it is unnecessarily restrictive.
- y y max and y y mln represent the upper and lower limits on the vehicle lateral position (y) as illustrated in Fig. 9b.
- the driveable corridor 908b is between these limiting curves. These limits exclude more than simply off-road/out-of-lane regions from the navigable corridor — they also extend to stationary and/or moving hazards in the roadway such as debris, pedestrians 906b or other vehicles 904b. Thus, a hazard in the roadway looks to the controller like a constriction in the corridor as illustrated by the arrows C in Fig. 9b.
- the host vehicle 902b is permitted to travel anywhere within the corridor 908b.
- Constraints can be softened by including the
- the MPC objective function can be configured to force the constrained optimal solutions to satisfy corridor constraints before minimizing front wheel sideslip.
- This hierarchy of objectives is achieved by setting constraint violation weights (p ⁇ ) significantly higher than the competing minimization weight (R aa ) on front slip. Then when constraints are not active, front wheel sideslip — and the corresponding threat — remains low.
- p ⁇ constraint violation weights
- R aa competing minimization weight
- Additional norms that may be used include Root Mean Square of predicted state or states over the prediction horizon, and any of the above mentioned norms, with weighting profiles, over the prediction horizon (i.e. state predictions at a chosen time, for
- closer to the current vehicle state may be weighted more heavily in the aggregate metric than predicted states at later times or at times other than the specifically chosen time.
- the chosen time may be other than the current time, such as a time immediately after some other event).
- front wheel slip is directly tied to, and tends to be a good indicator of, vehicle stability and controllability by front wheel steering.
- available surface friction places a measureable limit on how large front wheel slip angles can become before loss of control is imminent. This limit provides a useful benchmark against which threat
- assessments can be compared to assess maneuver stability (or nearness to instability).
- the path prediction explicitly minimizes the very metric used to assess threat. This hierarchy of objectives — remain within the corridor while minimizing front slip as much as possible — thereby provides a "best case” or minimal-threat assessment from a dynamically-feasible maneuver.
- the controller may not completely satisfy position constraints, making a an
- the cost-based prediction J SI is related to the predicted front wheel sideslip by
- a crowded location presents a more severe situation than traveling along an open road.
- the system may note this and invoke a different level of threat assessment, because of the complexity, or severity of the situation.
- the trajectory predictor (and actuator command generator in the case of semi-autonomous assistance) minimizes the very characteristics that describe threat. These characteristics, in turn, are based on an accurate model of the vehicle whose performance, stability, and safety may be assessed against an objective, physics-based metric.
- the threat assessor combines states, inputs, and constraint violations using a weighting function, that is of the same form and order as the weighting function used by the trajectory planner and actuator command generator.
- underground, etc. can be used with heavy equipment, such as agricultural equipment, fork lift trucks, cranes, etc., which not only move from place to place in an environment, but also, or alternatively, change their configuration from one moment to the next, such as when a crane is extending its boom. They may be used with naval vessels. Additionally, they may be used with unmanned or remotely-operated vehicles such as unmanned aerial drones, unmanned ground robots, etc.
- the methods and apparati may be used with surgical machines, which are guided by a human surgeon who manipulates an input device, to drive an output device, such as a scalpel, cauterization tool, stitching machine, etc.
- the environment of the surgical tool may be considered to include the body being operated upon, instrument supports, etc. They may also be used to control chemical processes, and power plants, such as nuclear, electric, etc.
- inventions disclosed herein are also useful with respect to devices that do not move from one point to another in an environment, and are also useful in connection with large systems, such as power plants, and also processes, such as chemical processes.
- a system that does not move within an environment still needs to account for the influence of external processes and disturbances upon it. Although these are not obstacles per se, they still represent threats to system success. In a chemical process, such an obstacle-like threat could be an external agency that dumps chemicals at different intervals into the mixture under control, or an uncontrolled temperature variation in the environment of the plant. Regarding a
- the term process plant is used herein and in the claims to refer to the volume of material that is being transformed and manipulated according to a chemical process, which may be in a stationary vessel, or moving along in a series of pipes and vessels, over time.
- the environment in which the process plant operates would include the vessels and atmospheres in which the process plant exists.
- Environmental disturbances, or hazards could include ambient temperature influences, chemical pollutants, vibrations of equipment, etc.
- inventions disclosed herein use the coordinates of such influences, in whatever coordinate system is relevant for understanding and evaluating their effect upon the process under control.
- the coordinates of chemical process hazards may relate to which vessel, or run of conduit a disturbance affects.
- Autonomous operation may be local or remote.
- Remotely operated systems typically exhibit a time lag. Because this method for threat assessment and semi- autonomous operator assistance is based on a model predictive control prediction, it is particularly well suited for
- Unmanned Ground Vehicles UUV s
- Unmanned Aerial Vehicles UAV s
- Unmanned Underwater Vehicles UUVs
- remote surgical equipment In relaying signals between the operator and the vehicle, such systems frequently experience
- the framework described here may be configured to explicitly incorporate these time delays into the optimal trajectory prediction, thereby eliminating the need for additional modules.
- Known systems may be illustrated with reference to Fig. 10.
- Known systems use distinct and sometimes competing modules to assist an operator.
- Such distinct, independent systems 1000 can include, but are not limited to: warning devices 1002, a yaw stability controller 1004; a roll
- the roll stability controller may output a steering command that is inconsistent with a steering command generated by the lane assist
- the inventions described here may operate as a unitary, consistent whole, 1100, in conjunction with, or independent of, the human operator.
- the MPC constitutes a state trajectory and control input planner 1102, which generates 1102 an optimal trajectory for the device, from current position through a time horizon, together with the device control inputs necessary to follow that trajectory.
- predicting the threat posed to the vehicle given the driver's current and past performance, current and past vehicle state evolution, and environmental features/constraints. It is physically accurate because it is based on a forward-simulated vehicle model. It is predictive, because the vehicle model is simulated over a future time horizon. It is flexible because threat and intervention decisions are based on a corridor — rather than path characteristics.
- Warning and operator feedback systems 1104 may inform any of a number of operator warning devices, including audible, visual, haptic, olfactory, etc.
- the inventions may be configured to predictively and semi-autonomousIy act as a device stability controller 1106.
- stability controllers that this system might replace include, but are not limited to yaw stability controllers, roll stability controllers, anti-lock brakes and traction controls.
- this framework predicts vehicle state evolution over a projected time horizon and adjusts the current vehicle state to keep both current and predicted stability-critical states (such as yaw angle, wheel slip angles, etc.) below critical levels.
- inventions are capable of is passive driver assistance 1108. These subsystems can apply various degrees of resistance, such as to steering wheel resistance torque overlays, braking (anti-lock brakes) traction control, yaw stability control, acceleration inhibitors, to discourage the operator from further increasing threat. (Note that there is some overlap between passive and reactive assistance systems and stability controllers . )
- Active intervention modes 1110 go beyond providing resistive feedback and actively initiate steering and/or braking commands .
- the inventions are well-suited to semi-autonomousIy assist the driver; utilizing sensory information related to the vehicle surroundings and a prediction of a safe vehicle trajectory through those surroundings to exert appropriate actuator effort and avoid impending accidents.
- Sensory information would include data related to nearby vehicles, pedestrians, road edges, and other salient features to assess accident threat.
- a second key advantage is that the invention may be configured to operate only during instances of
- the corridor-based threat assessment upon which this intervention is based presents a key advantage over existing semi-autonomous safety systems in that using a corridor (as opposed to a path) allows intervention that does not
- the inventions allow for human-controlled, computer-controlled, and shared human/computer vehicle operation.
- intervention law can also be defined to adapt to driver performance based on an assessment of driver skill, and/or to include considerations for driver preference, environmental conditions, previous threat metric values, previous control inputs, and other factors.
- the intervention law can also take on dynamics of its own, exhibiting, for example, hysteretic and higher-order behavior as a function of past, current, and predictive threat. See Fig. 1.
- the inventions are capable of fully- autonomous vehicle control. That is, in high-threat scenarios or when desired by the human operator, this system can take full control of the vehicle, safely navigating it through the environment while avoiding both collisions and loss of control .
- This multiplicity of highly competent options enables a unitary, integrated, mutually consistent, optimal operator assistance system. Because the modules for
- the system can allow and direct control to flow seamlessly from a mere warning, to passive assistance, stability control, active assistance, semi-autonomous and fully autonomous, and back again, over and over again, automatically. All of the threat response activities are mutually consistent. This permits a truly, reliably user configurable semi-autonomous system. All of this flexibility flows from the use of generating an appropriate and active control input based on predicted threat.
- the operations may be performed sequentially or simultaneously.
- controller and threat assessor may have a driver input.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Geometry (AREA)
- Business, Economics & Management (AREA)
- Educational Administration (AREA)
- Educational Technology (AREA)
- Aviation & Aerospace Engineering (AREA)
- Computer Hardware Design (AREA)
- Evolutionary Computation (AREA)
- General Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Entrepreneurship & Innovation (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Control Of Driving Devices And Active Controlling Of Vehicle (AREA)
- Traffic Control Systems (AREA)
Abstract
Various types and levels of operator assistance are performed within a unified, configurable framework. A model of the device with a model of the environment and the current state of the device and the environment are used to iteratively generate a sequence of optimal device control inputs that, when applied to a model of the device, generate an optimal device trajectory through a constraint-bounded corridor or region within the state space. This optimal trajectory and the sequence of device control inputs that generates it is used to generate a threat assessment metric. An appropriate type and level of operator assistance is generated based on this threat assessment. Operator assistance modes include warnings, decision support, operator feedback, vehicle stability control, and autonomous or semi-autonomous hazard avoidance. The responses generated by each assistance mode are mutually consistent because they are generated using the same optimal trajectory.
Description
AN INTEGRATED FRAMEWORK FOR VEHICLE OPERATOR ASSISTANCE BASED ON A TRAJECTORY PREDICTION AND THREAT ASSESSMENT
RELATED DOCUMENT
[0001] Priority is hereby claimed to: U.S. Provisional application No. 61/270,933, entitled SEMI-AUTONOMOUS CONTROL OF PASSENGER VEHICLES, in the names of Sterling J. Anderson, Steven C. Peters and Karl D. Iagnemma, filed on July 15, 2009, referred to herein below as the Comprehensive Provisional patent application, and U.S. Patent Application Serial No. 12/711,935, PREDICTIVE SEMI-AUTONOMOUS VEHICLE NAVIGATION SYSTEM, in the names of Sterling J. Anderson, Steven C. Peters and Karl D. Iagnemma, filed on February 24, 2010, and
International Application No. PCT/US2010/025206, PREDICTIVE SEMI-AUTONOMOUS VEHICLE NAVIGATION SYSTEM, in the names of Sterling J. Anderson, Steven C. Peters and Karl D. Iagnemma, filed on February 24, 2010, designating the United States, both of which claim priority to U.S. Provisional application No. 61/209,250, entitled PREDICTIVE SEMI-AUTONOMOUS VEHICLE NAVIGATION SYSTEM, in the names of Sterling J. Anderson,
Steven C. Peters and Karl D. Iagnemma, filed on March 5, 2009, priority to which is also claimed herein, referred to below as the Semi-Autonomous patent applications, which are all hereby fully incorporated herein by reference. This case is also related to a PCT application that is being filed on the same date, herewith, in the names of Sterling J. Anderson and
Steven C. Peters, and Karl D. Iagnemma, submitted by
electronic filing under attorney docket number MIT 13304C1PCT,
entitled, METHODS AND APPARATI FOR PREDICTING AND QUANTIFYING THREAT BEING EXPERIENCED BY A MODELED SYSTEM, which is hereby fully incorporated herein by reference, and is referred to herein below as the co-pending Threat Assessment application.
Background
[0002] Inventions described herein relate to a novel, corridor-based framework that performs threat assessment and provides varying degrees of mutually consistent automated operator assistance as a threat response in human-machine systems, such as locally or remotely-operated passenger vehicles, transport vehicles, agricultural machinery, fork lift trucks, aerial vehicles, robots, or surgical tools. This framework explicitly considers human and machine dynamics without presuming operator intentions or limiting the
avoidance maneuver (and its associated threat assessment) to a specific path. It provides a unified framework that allows for various modes and levels of mutually consistent operator assistance, from operator warning to stability control to passive intervention, to active semi-autonomous control, and finally, to autonomous machine operation.
[0003] Automotive active safety systems are concerned with preventing accidents through the introduction of various computer-controlled actuation methods to inform, improve, or override a human operator's steering and/or braking
performance. Active safety systems currently in existence include yaw stability control, roll stability control, traction control, and antilock braking, among others. While these systems reduce accident frequency, their path-based and largely reactive nature limits their ability to: 1) accurately assess the threat posed by a given scenario and 2 ) adequately determine when and how to intervene to assist the driver. This dependence on a specific trajectory (amidst a myriad of options available to the operator) reduces the accuracy and significance of the threat assessment and leads to controllers that selectively replace (rather than assist) the driver in order to follow an automation-designated path.
[0004] The inventions described and claimed herein relate, primarily, to the unified nature of the framework, and its flexibility in being able to accurately assess threat, and then participate in one or more of a wide variety of mutually consistent operator assistance modes, of varying levels of operator autonomy, from complete autonomy, to no autonomy.. Inventions described and claimed in the Threat assessment application relate primarily to threat assessment aspects of this novel framework. Both have similar disclosures.
[0005] A basic premise of threat assessment for such assisted and automated systems is generally as follows. First, sensing systems such as radar, LIDAR, cameras, inertial measurement units and GPS localization systems are used to detect, classify, and track the position of objects and the drivable road surface in the host vehicle's vicinity as well as measure vehicle states. Once these potential hazards have been identified, localized and their motion has been
estimated, a threat metric is used to quantify the threat they pose to the host vehicle, together with the threat of
departing the drivable road surface due to loss of vehicle control. As used herein, threat assessment or threat
prediction is used to mean identifying hazards and quantifying threat. Many threat assessment technologies are designed to then trigger and/or implement countermeasures to reduce the threat. These countermeasures can be passive or active. The effectiveness of threat assessment metrics depends on their ability to correctly identify hazards and accurately assess the threat that potential hazards pose to the host vehicle.
[0006] Threat metrics described in the literature
predominantly use time-based, distance-based, and
deceleration-based measures to characterize the threat level of a given scenario. Time-based threat measures project time to collision (TTC) based on current speeds, positions,
trajectories, and (in some formulations) other vehicle states. Distance-based metrics are generally calculated using
prevailing range and vehicle speeds and require constant velocity/acceleration assumptions and simple hazard geometry.
Finally, acceleration-based metrics assess the threat of a given maneuver based on the minimum (and often assumed
constant) lateral or longitudinal acceleration that a simple avoidance maneuver would require, given the current position, velocity, and acceleration of both host and hazard. In another approach, estimate is made of the lateral acceleration
required to execute a constant radius evasive maneuver. That implementation then compares this acceleration to a threshold value. When the required acceleration reaches this threshold, braking countermeasures are implemented to reduce the
vehicle's longitudinal velocity.
[0007] While the above threat metrics have been shown to provide useful estimates of the danger posed by a given maneuver, they suffer from many drawbacks. They are not well suited to consider multiple hazards, complex vehicle dynamics, or complicated environmental geometry with its attendant constraints. The geometrically-simple (straight-line or constant-radius-turn (CRT)) avoidance maneuvers assumed by these metrics may also misestimate the true threat posed by scenarios where the optimal avoidance trajectory follows a curve of varying radius or non-constant velocity/acceleration.
[0008] At least one known method of assessing threat relates to a vehicle that is intended to navigate along a path. The path may be predetermined, or calculated, based on data, such as information about obstacles and a path followed by a track, such as a roadway in the case of a road vehicle, such as an automobile. The path is a curve of simple
geometry, having essentially no width. If the vehicle
deviates from the zero width path, the system determines that danger has arisen, and the system generates a threat signal. However, in fact, the threat of actual danger is potentially low, because vehicles driven by a human operator typically operate within a field of safe travel, or a corridor, rather than along a relatively arbitrary line, such as the centerline of a roadway. This dependence on a specific trajectory (amidst a myriad of options available to the operator) reduces the accuracy and significance of the threat assessment and leads
to controllers that selectively replace (rather than assist) the driver in order to follow an automation-designated path.
[0009] The following terms will be used herein as follows. A path is a simple geometric curve in two-dimensional space, along which a vehicle may travel. The path, in x-y space may be defined by a function, y=f(x). A path has a width of essentially zero. A trajectory is a physically-achievable and time-parameterized sequence of vehicle states (such as
velocity, yaw angle, wheel sideslip angle, etc.) over a time horizon, By physically-achievable, it is meant that for every trajectory, there exists a set of controller inputs, such as braking torque and steering angle, that when applied to a model of the vehicle produce the desired trajectory. It has been mentioned that the trajectory includes the velocity of a vehicle as an element. It may also be thought of as having time as a parameter of the path, which may then establish velocity at different locations.
[0010] A corridor is a swath through two-dimensional space, which may be defined by an inequality ymln < y < ymax, where each of ymln and ymax are themselves defined by ymax=g(x) and ymln=h(x). Thus, a corridor may be considered to be the space between two curves in two-dimensional space. Travel anywhere within the corridor is considered to be safe. A region is a concept that is defined in connection with inventions hereof, and it will be defined below.
[0011] A Model Predictive Controller (MPC) is an optimal control method typically used to generate an optimal set of control inputs (spanning through a future time horizon) required to track a desired path while minimizing a user- defined objective function. In typical usage, only the first element of this command input sequence is implemented at each solution timestep and the remainder are disregarded.
[0012] A human driver typically operates a vehicle within a safe range of vehicle states. For example, the driver
typically maintains the vehicle's position on the roadway
(lateral position state) within the corridor defined by road or lane edges. Similarly, other states such as vehicle velocity, lateral acceleration, etc. provide some indication of threat to the driver, who (consciously or subconsciously) seeks to keep them within a safe operating range (or between upper and lower bounds). Depending on the driving conditions, posted speed limits, and other factors, for example, a driver may allow vehicle velocity to vary between 53 and 70 miles per hour rather than slavishly maintaining 55 miles per hour along a roadway. Likewise with vehicle sideslip, which the driver will typically strive to maintain within a reasonable (or safe) range. Thus, a human operator operates within an tridimensional region of state space, rather than along a simple, zero, or nearly-zero width curve of a physical trajectory.
[0013] In many cases, it may be desirable to assess threat based on a realistic roadway corridor than an unrealistic single path on the roadway. Similarly, it may also be
desirable to assess threat based on a realistic N-dimensional region of state space (as explained more fully below, which includes as a portion, the two-dimensional corridor) .
[0014] Some known threat assessors exist as a separate system, not integrated with other systems of the device analysis and control apparatus. These systems may base the assessment of threat on a device state exceeding a relatively arbitrary threshold. Further, these threat assessors
typically do not provide the control input necessary to decrease the threat they've assessed.
[0015] Other threat assessment approaches use only
rudimentary threat assessment metrics based on, for example, the current deviation of the device from a predetermined optimal path. It would be desirable to be able to take
advantage of predicted future states of the device in
assessing threat.
[0016] Still other threat assessment approaches also do not consider combined effects of vehicle dynamics, stability
constraints and terrain interactions to evaluate maneuver severity. However, it would be desirable to be able to
consider these matters in threat assessment.
[0017] Thus, an object of inventions disclosed in the
Threat Assessment application is to take advantage of
predicted future states and predicted future optimal inputs to assess threat. A further object would be to be able to
consider combined effects of vehicle dynamics, stability constraints and terrain interactions to evaluate maneuver severity in assessing threat. Still another object would be to be able to assess threat on a realistic corridor, rather than an unrealistic single path. Another object would be to assess threat with an apparatus that is integrated with other systems of system analysis and control, which are also used to control the device or system, rather than assessing threat on a device state exceeding an arbitrary threshold. Still another object would be to use the assessment of threat to generate an operator assistance signal to assist the operator in safely operating the device or replace the operator as necessary to ensure safe operation of the device.
[0018] Known device control systems use distinct and sometimes competing modules to assist an operator. Thus, they often provide conflicting and inconsistent operator assistance threat responses.
[0019] These and other objects of inventions disclosed herein will be more fully explained and understood with reference to the Figures of the Drawing, which are:
BRIEF DESCRIPTION OF THE FIGURES OF THE DRAWING
[0020] Fig. 1 is a block diagram illustrating basic
framework operation;
[0021] Fig. 2 graphically shows an example of various potential intervention laws based on threat metric
calculation;
[0022] Figs. 3a and 3b graphically show an obstacle
avoidance scenario illustrating different stages of
intervention for an inattentive driver, with Fig. 3a showing the different levels of intervention, and Fig. 3b showing the relative locations of the host vehicle and the environment;
[0023] Fig. 4 shows, in flowchart form, basic steps of a method of threat assessment and semi-autonomous control, with possible considerations at each step;
[0024] Figs. 5a, 5b and 5c show, graphically, a simulated test illustrating system response when driver fails to
navigate a curve in the road, as shown schematically in Fig. 5a, with steering angle shown in Fig. 5b and control authority K shown in Fig. 5c, where K represents the proportion of control authority given to autonomous system, with the driver allowed the remaining (1-K);
[0025] Figs. 6a, 6b and 6c show, graphically, a simulated test illustrating system response to an erroneous driver swerve, as shown schematically in Fig. 6a, with steering angle shown in Fig. 6b and control authority K shown in Fig. 6c, where K represents the proportion of control authority given to autonomous system, with the driver allowed the remaining (1-K);
[0026] Figs. 7a, 7b and 7c show, graphically, a simulated test illustrating system response when driver fails to
anticipate/avoid obstacle, as shown schematically in Fig. 7a, with steering angle shown in Fig. 7b and control authority K shown in Fig. 7c, where K represents the proportion of control authority given to autonomous system, with the driver allowed the remaining (1-K);
[0027] Fig. 8 shows, in flow chart form, steps of a method of an invention hereof for assessing threat, with additional considerations associated with some of the steps;
[0028] Fig. 9a shows, schematically, an illustration of a simple path tracking control set up of the prior art;
[0029] Fig. 9b shows, schematically, a corridor keeping control set up, using position constraint y=f(x) for static hazards;
[0030] Fig. 10 is a schematic representation in block diagram form illustrating how a representative system of the prior art that can lead to poor, mutually inconsistent
performance;
[0031] Fig. 11 is a schematic representation in block diagram form showing an array of functions performed in a unified, mutually consistent manner by inventions hereof.
DETAILED DESCRIPTION
[0032] Inventions described in the above-referenced Semi- Autonomous applications, and the Comprehensive Provisional patent application, relate, among other things, to a unified framework for performing threat assessment and semi-autonomous vehicle navigation and control while allowing for adaptable and configurable intervention laws and configurable control inputs .
[0033] Inventions described in the Threat Assessment application, and also below, relate to methods and apparati for identifying and quantifying threat being experienced by a system that can be modeled, such as a vehicle, such as a road vehicle, such as an automobile.
[0034] To summarize briefly inventions hereof, various types and levels of operator assistance are performed within a unified, configurable framework. A model of the device with a model of the environment and the current state of the device and the environment are used to iteratively generate a
sequence of optimal device control inputs that, when applied to a model of the device, generate an optimal device
trajectory through a constraint-bounded corridor or region within the state space. This optimal trajectory and the sequence of device control inputs that generates it is used to generate a threat assessment metric. An appropriate type and
level of operator assistance is generated based on this threat assessment. Operator assistance modes include warnings, decision support, operator feedback, vehicle stability
control, and autonomous or semi-autonomous hazard avoidance. The responses generated by each assistance mode are mutually consistent because they are generated using the same optimal trajectory.
[0035] As used herein, identifying and quantifying threat will be referred to as assessing threat, or sometimes,
predicting threat. Danger may also be used interchangeably with threat. The methods and apparati for assessing threat are of the same type as can be used in the unified framework for semi-autonomous vehicle navigation and control, discussed in the applications mentioned above. Such methods and apparati can be used for assessing threat in any system that can be modeled.
[0036] In contrast to existing approaches, the inventions described herein do not rely on a subjective prediction of a path intended by a human operator or a specific path proposed by an automatic path planning algorithm. Instead, these inventions base threat assessment and operator assistance decisions on dynamic properties and known constraints inherent to the vehicle and the environment. These inventions warn of deviation from a physically-constrained and dynamically- feasible region in N space (that includes a two-dimensional corridor in physical space) rather than an arbitrarily- calculated path, which is often less meaningful since it often does not represent a true, or even achievable, much less optimal route. Further, by predictively simulating the vehicle dynamics over a time horizon, which may be finite or infinite, its threat assessment and intervention operations explicitly and pre-emptively consider the combined effects of vehicle dynamics, stability constraints, and terrain interaction on maneuver severity.
[0037] Turning again to definitions of some helpful
terminology, a region is an N-dimensional analogue of a
corridor. A region is an n-dimensional area defined in the state space of the device model over a time horizon. This region is bounded by corridor constraints (which apply to vehicle position states), together with other state
constraints (such as those imposed on vehicle states such as yaw angle, yaw rate, velocity, wheel sideslip angle, etc.).
[0038] In contrast to known MPC planners, MPC planners of inventions hereof are not required to utilize a single
reference path for optimal control input generation. Rather, the MPC planners of inventions hereof are capable of
generating an optimal trajectory that constitutes a path through a constraint-bounded corridor or region of the state space. Thus, planners described herein generate a course of motion that need not follow a predefined path but may instead generate its own optimal trajectory (and the control inputs necessary to achieve it) at successive time steps.
[0039] Threat assessment inventions described herein may be used with any system that can be modeled, including vehicles, such as terrestrial, nautical, and aerial vehicles,
manufacturing machines, chemical processes, and fully
automated processes, such as nuclear power plant operations and economic processes. All such systems and devices that can be reasonably modeled, can have associated therewith
conditions that threaten operator, client, or third party use and enjoyment or process effectiveness. As such, there is benefit in assessing these threatening conditions, and issuing warnings therefore, possibly enabling counter measures or other preparations. The specific implementations described herein are illustrated in the context of a terrestrial
vehicle, such as an automobile. However, it is to be
understood that the explanation is for illustrative purposes only, and that the inventions described herein can be used to identify and quantify (i.e., assess) threat in all systems that can be modeled. The systems may be physical, such as devices and chemical processes. They may also be non-physical, such as economic systems. In all cases, the system model will include states (such as pitch angle for an aerial vehicle,
solvent concentration for a chemical system, or asset prices for a financial system) for which a desired value or range of values exists and to which constraints (such as stall limits for an aerial vehicle, saturation for a chemical process, or price caps for a financial system) may apply.
[0040] Turning then to terrestrial systems, for
illustration purposes only, automotive active safety systems are concerned with preventing accidents through the
introduction of various computer-controlled actuation methods to improve driver braking and steering performance. Current active safety systems include yaw stability control, roll stability control, traction control, and antilock braking, among others. While these systems reduce accident frequency, they are fundamentally reactive in nature: their intervention is based on current vehicle (and, possibly, road surface) conditions. Because they do not utilize 1) sensory
information related to the vehicle surroundings or 2 ) a prediction of the vehicle's path through its surroundings, they have limited ability to assess the threat of impending accidents, and thus cannot exert corrective actions to avoid them.
[0041] Active navigation systems aim to avoid accidents by utilizing sensory information related to the vehicle
surroundings and a prediction of a safe vehicle trajectory through those surroundings to exert appropriate actuator effort to avoid impending accidents. Sensory information would include data related to nearby vehicles, pedestrians, road edges, and other salient features to assess accident threat.
[0042] Except in cases of desired full automation, such navigation systems ideally operate only during instances of significant threat: it should give a driver full control of the vehicle in low threat situations but apply appropriate levels of computer-controlled actuator effort during high threat situations. An active navigation system can therefore be termed semi-autonomous, since it must allow for human-
controlled, computer-controlled, and shared human/computer vehicle operation. Such a system should be as unobtrusive to the driver as possible (i.e. it should intervene only as much as is minimally required to avoid an impending accident).
[0043] A semi-autonomous active navigation system described in part in the above referenced Semi-Autonomous patent
applications can satisfy the above requirements and desired characteristics. Further, it provides a framework into which various distinct sensing and actuation modes can be easily incorporated. The system's method for threat assessment and computer-controlled intervention can potentially be modified in real time based on the scenario, environmental conditions, driver preference, or past driver performance. Fig. 1 shows, schematically, in block diagram form, a basic framework operation.
[0044] A model Predictive Control (MPC) vehicle navigation block 110 starts with a model 112 of the environment, a model 114 of the device, in this case, a vehicle, and the vehicle's current state, including the position. It generates an optimal vehicle trajectory from the current position through a time horizon. The trajectory is optimal with respect to a predefined, configurable set of criteria. It also generates a corresponding optimal set of control input commands necessary to execute an optimal trajectory within the corridor and ensure that the vehicle 102 operates within safe driving limits (defined by a constraint-bounded region in the state space). The environment model can be based on a priori known information (e.g. from maps) and/or information gathered by real time sensors, such as on-vehicle sensors 104 (e.g.
cameras and laser rangefinders, vehicle to vehicle (V2V sensors), and can include information 106 related to the environmental and potential environmental hazards, such as position of road edges, lane boundaries, holes, slopes, static obstacles (e.g. trees, road-side signs), and dynamic obstacles (e.g. other vehicles, pedestrians). The vehicle model is user-defined and can be of varying complexity and fidelity.
The real-time sensors may also be mounted in the environment and communicate with the control system on the vehicle.
[0045] The predicted safe vehicle trajectory (and
associated control inputs to yield such a trajectory) is generated such that it satisfies a configurable set of
requirements, including, for example, that the vehicle
position remain within a safe driving corridor of the roadway, that the vehicle sideslip angle not exceed the safe limit of vehicle handling, that tire friction forces not exceed a surface friction-limited value, and others. Note that by permitting the trajectory to satisfy locations within a corridor, rather than the more stringent, yet more arbitrary, locations along a (essentially zero width) path, the
controller avoids restricting its solution to one particular path and instead iteratively calculates the optimal trajectory within the corridor at every sampling time step. The control inputs can be associated with one or multiple actuators, such as active steering, active braking, and others. The predicted vehicle trajectory and associated control inputs may be generated via constrained optimal control, which leverages efficient optimization methods and constraint-handling
capabilities. In particular, model predictive control
techniques may be used to generate a set of optimal
trajectories and associated control inputs, etc. It should be understood that trajectory is used herein to mean a sequence of vehicle states, including its position, velocity, sideslip and yaw angles, etc.
[0046] At successive discrete sampling instants, the predicted vehicle trajectory and predicted control inputs are analyzed by a threat assessor 108 to quantify the threat to the vehicle by computing a configurable metric, such as the maximum lateral acceleration, sideslip angle, or roll angle over the trajectory, the minimum proximity to obstacles, or other metrics. Generation and use of this threat assessment metric is discussed in more detail below, and is a focus of inventions described herein. Threat may be more formally considered to be a hierarchical combination of obstacle
avoidance, stability-critical states, inputs, etc, based on a model of the vehicle.
[0047] In the case of a semi-autonomous system, the control authority exerted by the system can then be determined as a function of this generated threat: generally speaking, if the threat metric value is low, the control system intervention is low (i.e. the driver commands the vehicle with little or no computer-controlled intervention); if the threat metric value is high, the control system intervention is high. The form of the intervention law modulating this control system authority is configurable and can differ for different actuators (i.e. a vehicle with both active steering and braking can have
distinct intervention laws defined for the steering actuator and the braking actuators). The intervention law can also be defined to adapt to driver performance based on an assessment of driver skill, and/or to include considerations for driver preference, environmental conditions, previous threat metric values, previous control inputs, and other factors. Fig. 2 shows, schematically, examples of various potential
intervention laws, showing, from top to bottom, linear, smooth and threshold-shaped intervention laws that depend only on predicted threat. The vertical axis represents the degree of control authority given to the active navigation and control system while the horizontal axis represents the predicted threat, with cause for intervention increasing from left to right.
[0048] In the system described above, as the threat metric value increases, indicating that the predicted vehicle
trajectory will near a pre-defined critical vehicle state(s) (such as spatial location, lateral acceleration, or tire friction saturation), the control system begins to assume control authority to preempt an unsafe maneuver. As the threat metric decreases, the controller's authority phases out. In this manner, the system can be said to be both predictive and semi-autonomous .
[0049] Note that in extreme cases, when the driver does not perform an appropriate corrective action, it is conceivable that a required hazard avoidance maneuver will reach vehicle handling limits. To account for such scenarios, the
intervention law can be designed such that it assumes full authority by the time the predicted safe trajectory reaches the limit of any pre-defined critical vehicle states. This corresponds to a situation where only an optimal set of inputs would result in a safe vehicle trajectory.
[0050] Fig. 3 shows schematically an obstacle avoidance scenario illustrating different stages of intervention for an inattentive driver. Initially, the host automobile 302 is at location 1. No obstacles are in view, and the optimal
predicted trajectory is a straight path. The predicted threat is at a low level, indicated by the vertical line designated 1, near to the left hand side, which represents low threat. As the vehicle 302 advances along the roadway to the location 2, it comes nearer to a truck 304, whose velocity is either zero, or much less than that of the host vehicle 302. The sensors sense the proximity of the obstacle vehicle 304, and generate a threat metric that is larger, as at the vertical line designated 2, near the right hand limit of the threat scale. The optimal predicted trajectory assumes a curved shape, to avoid the obstacle 304. Simultaneously, the level of
intervention K, increases (as shown by each of the three different intervention laws) so that, in a semi-autonomous system, the controller would take more and more control, the nearer to the obstacle the host vehicle 302 comes. In other systems, the controller may take different countermeasures, such as initiating a warning, priming brakes, seatbelts, or airbags, and/or engaging active systems, etc.
[0051] Fig. 4 shows, schematically, in flow chart form, a basic flow of steps performed by a controller of an invention described in the Semi-Autonomous applications, above, with possible considerations at each step.
[0052] An initial step 402 generates an optimal set of control inputs and corresponding vehicle trajectory by forward simulation. Considerations 402a for this step include, for example, (but are not limited to) the vehicle dynamics, current state of the vehicle and environment, terrain and environmental disturbances, available actuation, trajectory objectives, safety limits, and driver inputs.
[0053] A next step is to assess 404 the predicted threat to the vehicle. General considerations 404a for this step include characteristics of the optimal path and associated control input, safety limits and driver inputs. The method of threat assessment is discussed below in more detail in connection with Fig. 8.
[0054] Returning to a brief discussion of a semi-autonomous system, a next step is to generate D06 control authority gains, with a major consideration D06a at this stage being the desired intervention characteristic. The next step D08 is to implement the scaled control for the current time.
[0055] Simulation studies have been conducted. They are presented here to show how the threat assessment metric might be used to govern the level of assistance provided to a human operator. Figs. 5a, 5b and 5c show, graphically, the results of a simulated test illustrating system response when a driver fails to navigate a curve in the road, shown by in Fig. 5a by a pair of lines. The trajectory that the driver would have followed without assistance is shown dashed. Note that it leaves the roadway. With assistance, it is shown solid black and remains within the roadway. Note that here, K represents proportion of control authority given to the autonomous system, with the driver allowed the remaining (1-K). The middle graph, Fig. 5b shows the steer inputs, with the dashed line corresponding to the driver and the solid curve
corresponding to the control system. The lower graph, Fig. 5c, shows the control authority given to the autonomous system, in this case, steering, with the degree varying with distance (x) along the horizontal scale.
[0056] Figs. 6a, 6b and 6c show, graphically, the results of a simulated test illustrating the system response to an erroneous driver swerve. Again, K represents proportion of control authority given to autonomous system, with the driver allowed the remaining (1-K). The same line types as above correspond to the driver without assistance (gray dashed) and with assistance (solid line). The safe roadway corridor is shown in Fig. 5a by a pair of light solid lines in the upper graph. Distance is shown along the horizontal scale. The assisted trajectory remains within the safe roadway.
[0057] Figs. 7a, 7b and 7c show, graphically, a simulated test illustrating system response when a driver fails to anticipate/avoid an obstacle, similar to the scenario
illustrated above with respect to Figs. 3a and 3b. Again, K represents the proportion of control authority given to autonomous system. The obstacle is simulated by a jog in the light line that represents one edge of the safe roadway. The only inputs used in this simulation are, again, steering of the driver and the autonomous system.
[0058] Significant advantages stem from the predictive nature of this solution. In addition to considering past and current vehicle and driver actions to assess threat and determine control authority, the solution generated by the present inventions predicts a future vehicle trajectory and associated threat, and uses this prediction to schedule control authority.
[0059] This predictive nature also allows for a more accurate assessment of threat than is otherwise possible.
While other threat assessment metrics rely largely on physics- based calculations, the metrics used in the inventions
disclosed herein can derive from sophisticated physics based vehicle and environmental models. These models yield very accurate threat assessments by considering the effects of terrain conditions, environmental disturbances, and physical limitations of vehicle actuators. These models can also
assess threat for more complex vehicle trajectories than is possible with simplified models.
[0060] Threat assessment techniques disclosed herein, and uses within a system as described herein provide improved modularity and adaptability when compared to previous methods and apparati. Aspects of this improved modularity are
discussed in more detail below. The underlying control framework can accommodate multiple actuation modes and vehicle models, allowing for ready application of the system to various vehicle types and actuator configurations. The system's intervention law is also readily adapted (i.e. it can change over time based on an assessment of driver skill, driver preference, environmental conditions, previous threat metric values, previous control inputs, and other factors). These adaptations can be performed either statically or dynamically.
DETAILED DISCUSSION OF THREAT ASSESSEMENT
[0061] Turning now to a more detailed discussion of threat assessment techniques, first, the methods will be described generally, with the aid of a block diagram, and then a more detailed mathematical basis will be provided.
[0062] Gauging threat, using metrics based on predicted vehicle state evolution within a region/corridor is novel. This includes threat assessment based on a trajectory that remains within that corridor, along with semi-autonomous control necessary to keep the vehicle within the safe
corridor. Various threat assessment methods exist. Some are based on a predicted and/or optimal vehicle trajectory.
However, none use a region or corridor of safe travel to characterize the trajectory. By corridor, it is generally meant a portion of physical space, such as the width of a roadway, or a roadway and adjoining break down lane. By region, it is generally meant a region in N-space, in which two of these N dimensions may be the two spatial dimensions of the roadway's width and length, and other dimensions may be
states and/or control inputs of the vehicle, such as sideslip angle, yaw angle, velocity, steering angle, etc. This leads to significant difference in performance of the two general approaches. The simple path-based approach of the prior art is suited only to warn or selectively replace a human
operator. The corridor/region approach described herein may warn, supplement, or act in conjunction with a human operator, as well as replace.
[0063] Turning now to Fig. 8, a representative process for assessing threat is described. The steps about to be
described, 802-808, are all conducted at each time step of conducting a model predictive control operation. An optimal path is generated 802 through an environment, using model predictive control techniques. Considerations 802a that are taken in conjunction with this step may include, but are not limited to: vehicle dynamics and constraints, trajectory objectives (controller objective function), terrain and environmental disturbances and environmental information, such as: obstacle locations, terrain properties (slope, roughness, friction coefficient, etc.) and other disturbances (such as wind) . It is through these considerations of vehicle dynamics and constraints and obstacle locations, that the corridor aspects of this method of threat assessment enters. Trajectory objectives more broadly describes the penalties applied to violating constraints and/or approaching unstable states and shapes what an optimal solution looks like within the
corridor.
[0064] The model predictive controller generates an optimal trajectory, which constitutes an optimal sequence of inputs and the corresponding set of optimal vehicle states over a time horizon. These outputs are generated by the MPC unit, based on analysis of the models of the vehicle and the
environment, using model predictive control over a time horizon, as discussed in more detail below.
[0065] It is to be noted that threat assessment methods disclosed herein operate on an optimal (sometimes referred to
herein and the literature as best case, or benchmark)
trajectory that was generated to remain within a constraint bounded, traversable, corridor, while maximizing vehicle stability. Prior art path based methods use a desired
trajectory. The difference between optimal and desired, is that a desired path gives a rather inflexible and in some cases, arbitrary goal state to which the driver will be forced to adhere or risk setting off warning indicators or controller intervention, while the optimal trajectory gives an indication of the threat posed to the vehicle and a backup plan in case the human operator doesn't keep the vehicle within the
traversable corridor him/her self.
[0066] Consider a situation in which, for example, a driver diverges from a lane centerline (the desired path for prior- art systems). While a path based system will try and pull the driver back, inventions disclosed herein will re-compute a new trajectory that may be completely satisfied by staying off the centerline, as long as the vehicle is within the corridor.
[0067] The optimal trajectory, of predicted states and predicted vehicle control inputs, over a time horizon, is next coalesced 804 to generate a scalar prediction metric.
Considerations 804a that contribute to the step of generating a scalar metric may include but are not limited to: past, present and/or predicted states, inputs and objective function costs. Various norms (discussed below) may be used to combine trajectory components into this scalar. Past present and/or predicted operator inputs and performance may also, but need not be taken into account.
[0068] The result of the coalescing step 804 is a scalar prediction metric. This scalar can be non-dimensionalized 806 by normalizing it against known or approximated physical limits and/or predetermined desired thresholds to obtain a threat assessment at the current time. Suitable candidates for such thresholds include, but are not limited to maximum sideslip angle before loss of control, maximum load transfer before wheel liftoff, maximum lateral acceleration before
skidding, maximum longitudinal acceleration before skidding, maximum total acceleration before skidding, maximum steer angle before actuator saturation, maximum available
acceleration torque, and maximum available braking torque.
[0069] Considerations 806a that may be taken into account include, but are not limited to known physical limits on vehicle dynamics (e.g. friction saturation limits, rollover thresholds, etc), desired intervention characteristics, operator performance and operator preference.
[0070] The result of the non-dimensionalizing step 806 is a threat assessment, which can then be used for one or more of a variety of mutually consistent threat response functions, depending upon the system configuration. The threat assessment can be used to generate a warning, which may be perceptible by any human sense, including audible, visual, haptic and
olfactory. The threat assessment can also, or alternatively, be used to trigger assistance, which may be passive, or active, as discussed above, and, if active, to varying
degrees, depending on autonomy considerations. The threat assessment may even be used to modulate other system or vehicle characteristics.
[0071] Examples of passive assistance in the context of vehicular control include, but are not limited to: resistance torques on steering wheel, traction control, anti-lock braking systems. Examples of active assistance include, but are not limited to adaptive cruise control, yaw stability assistance Electronic Stability Control (ESC) lane keeping assistance, obstacle avoidance. Modulation of other system characteristics may include but are not limited to: seatbelt pretensioning, brake priming, suspension modifications (active suspensions, suspension stiffness modifications, etc.). Examples of semi- autonomous (a form of active assistance) assistance include, but are not limited to partial control of the steering and braking commands sent to the vehicle. Examples of autonomous assistance include, but are not limited to, full control of
one or more vehicle actuator such as wheel steering angles, braking torques, acceleration torques, etc.
[0072] Each of the foregoing steps, from the step 802 of generating an optimal path, through using 808 the threat assessment to initiate a threat response, such as to take some action, if need be, is conducted at each iteration of the Model Predictive Control operation, which repeats at a frequency tailored to the particular process under control. For instance, in a representative automobile control system, the MPC routine has been conducted 20 times per second, (each timestep requiring 50 millisecond to compute). Of course, different applications with different hardware and even different vehicle models have drastically different
parameters .
[0073] Turning now to more formal considerations, Model Predictive (MPC) (or receding horizon) Control is a family of finite-horizon optimal control schemes that iteratively minimizes a performance objective defined for a forward- simulated plant model subject to performance and input
constraint. Stated another way, MPC uses a model of the plant to predict future vehicle state evolution and optimize a set of plant control inputs such that this prediction satisfies constraints and minimizes a user-defined objective function.
[0074] Model predictive control has a number of significant properties that make it particularly well suited to threat assessment for use with, for instance, autonomous and semi- autonomous vehicle navigation problems. Its ability to
explicitly consider environmental, performance and actuator constraints enables corridor-based navigation and allows it to operate near the limits imposed by those constraints. This environmentally aware prediction, coupled with an objective- function-optimal control law, has been shown to closely mimic the performance of a human driver. Some implementations use a finite prediction horizon, which fits naturally with and may be based on the information provided by finite-horizon, forward-looking sensors. Additionally, the model-based nature
and multivariable-compatibility of the control calculation allows MPC to account for and easily adapt to structural changes and actuator availability from one vehicle model and/or loading configuration to the next. This adaptability may allow for reduced-cost controller implementation across product families and through ever-shifting safety
requirements .
[0075] Finally, MPCs predictive nature allows the
innovative use disclosed herein of certain constrained
configurations to automatically plan a path within a
partitioned environment without requiring any pre-defined vehicle trajectories. That is, where other control methods require a specific pre-planned path through the environment (which is often planned by a separate and suboptimal system) , MPC can be configured to according to inventions disclosed herein, to plan its own, optimal, path given a set of
situational position constraints. The path thus planned through the (pre-delineated) safe operating environment potentially offers a number of advantages over alternative trajectory planning methods; not only is it explicitly aware of vehicle dynamics, measured disturbances, and actuator limitations, but the constraint-satisfying trajectory plan it generates is feasible, since it is obtained from an already- generated set of control inputs. It is also optimal, with respect to some performance metric such as minimum lateral acceleration over a future time horizon, minimum wheel slip, etc. In the semi-autonomous framework described below, this optimal prediction can serve not only as a optimal trajectory plan, but also as an effective threat assessor.
[0076] At each time step, t, the current plant state is sampled and a cost-minimizing control sequence spanning from time t to the end of a control horizon of n sampling
intervals, t+nΔt, is computed subject to inequality
constraints (which establish the safe travel corridor). The first control element of this input sequence is implemented at the current time and the process is repeated at subsequent time steps. A state sequence spanning the same time period is
also generated. No known MPC process uses elements of either of these sequences, after it has identified the single control input element for the current time. Only that single, current control input element is used.
[0077] Inventions disclosed herein use these subsequent elements of one or both the state and control sequences to assess threat to which the device is (or may in the future be) exposed.
[0078] The optimal-control-based method that uses sensor information (and corresponding corridor boundaries) to
generate controller inputs also enables generation of a metric for analyzing the threat posed to the device by a given scenario. This metric is comparable to and in many situations more useful/accurate than existing metrics because it is based solely on known or approximated physical limits of the
environment and a (generally very accurate) model of the vehicle.
[0079] In contrast to known methods of threat assessment used in the prior art, threat assessment methods and apparati, as disclosed herein, generating instantaneous threat from an MPC-derived optimal avoidance trajectory inherently considers multiple hazards, actuator limitations/effects, measured disturbances, and (using nonlinear MPC), variable vehicle velocities and accelerations. Configuring the controller to plan a (sideslip-minimizing) trajectory within a safe region of travel ensures that the MPC-based threat assessment
provides a true assessment of the minimum instantaneous threat posed to the vehicle. In a driver warning context, threshold threat values may trigger driver warnings at critical/desired threat thresholds. For semi-autonomous control via computer control, threat assessment may be used to determine when and how strongly to intervene. The latter application is a topic of the Semi-Autonomous applications. This disclosure focuses on the design and development of the threat assessment metric itself .
[0080] A corridor-based trajectory-planning method may be used, based on constrained optimal control. When the objective function and constraints are defined as described below, the vehicle path calculated at each time step by the MPC
controller is assumed to be the best case or safest path within a corridor through the environment. Some key metrics from this prediction may be used to assess the instantaneous threat posed to the vehicle.
[0081] For a discrete plant model described by χ i+i = Axi + BΛ + BΛ
yk = Cx, + Dvv, with x, y, u, and v representing states, outputs, inputs, and disturbances of the system respectively, a quadratic
objective function over a prediction horizon of p sampling intervals is defined as
where Ry, Ru, and RΔU represent diagonal weighting matrices penalizing deviations from γ± = rlf and U1 = 0, pε represents the penalty on constraint violations and ε represents the maximum constraint violation over the prediction horizon p. Inequality constraints are defined as: y 7min (/ )- εV7min (z )≤ y J (k + ϊ + 1 I k)≤ y 7max (/ )+ εV7 max (/ )
U7min (/ )≤ U7 (k + i + 1 I k)≤ U7max (/ )
Δu7mm (i)≤ Δu7 (k + i + 1 1 k)≤ Δu7 max (/ )
i = 0,..., p - l
ε≥O where the vector Δu represents the change in input from one sampling instant to the next, the superscript "(9)3 "
represents the jth component of a vector, k represents the current time, and the notation (»)j(k+i|k) denotes the value
predicted for time k+i based on the information available at time k. The vector V allows for variable constraint softening over the prediction horizon, p, when ε is included in the objective function.
[0082] For reference trajectory tracking along a
geometrically simple path, lateral deviation of the vehicle's center of gravity (yy) from the corridor centerline (ry) is penalized by including Ry > 0 in an objective function of the form:
lc+p 1 k+p-l i k+p-l i
Λ = y -W - rj Ry Ky1 - rj+ > -U1 R11U1 + > -Au1 R411AM, where, k represents the current moment in time, p represents the number of time steps in the prediction horizon, Ry, Ru, and R^u represent weighting matrices on the lateral position state (y), inputs (u), and input rates (Δu) . The resulting
trajectory-tracking setup of a prior art controller and threat assessor, through a hazard-containing environment may then be illustrated by Fig. 9a.
[0083] Fig. 9a shows a vehicle 902a that seeks to avoid an obstacle of another vehicle 904a and a pedestrian 906a. A path based controller and threat assessor would attempt to follow a desired single track path ydes. This path may safely navigate the hazards, but it is unnecessarily restrictive.
[0084] For corridor-keeping, penalties on deviation from a desired trajectory (Ry) are replaced with lateral position constraints. This form of corridor-based navigation assumes that the environment has been delineated, with the boundaries of the navigable road surface at each time step described by the constraint vectors
In the preceding expression, yy max and yy mln represent the upper and lower limits on the vehicle lateral position (y) as illustrated in Fig. 9b. The driveable corridor 908b is between these limiting curves. These limits exclude more than simply off-road/out-of-lane regions from the navigable corridor — they also extend to stationary and/or moving hazards in the roadway such as debris, pedestrians 906b or other vehicles 904b. Thus, a hazard in the roadway looks to the controller like a constriction in the corridor as illustrated by the arrows C in Fig. 9b. The host vehicle 902b is permitted to travel anywhere within the corridor 908b.
[0085] For the constraint space to remain feasible v Jymax - vJγmi■n > 0u.•
[0086] Constraints can be softened by including the
magnitude of their violation ε in the objective function, which takes the form
k+p i k+p-l i k+p-l i i
J, = >— yη R v + > — U1 R w + > —AM, R »,,AW, +— pvε v , Rvv = 0 .
[0087] The MPC objective function can be configured to force the constrained optimal solutions to satisfy corridor constraints before minimizing front wheel sideslip. This hierarchy of objectives is achieved by setting constraint violation weights (pε) significantly higher than the competing minimization weight (Raa) on front slip. Then when constraints are not active, front wheel sideslip — and the corresponding threat — remains low. When the solution is constrained, predicted front wheel sideslip increases with the severity of the maneuver required to remain within the navigable corridor.
[0088] Physical limits on tire cornering friction dictate maximum safe angles of wheel sideslip. These angles provide an objective limit against which predicted sideslip may be normalized; when predicted threat approaches this known limit, loss of stability is imminent. This inherent limitation on
stability-critical states such as front wheel slip makes them particularly well suited as objective threat assessors.
[0089] Various norms may be used to reduce the vector of MPC predicted vehicle states x to a scalar threat metric Φx{k)
(instantaneous threat assessment at time k) . The performance of several norms has been studied. Table 1 describes how each was determined.
[0090] Table 1 Norms used to reduce MPC predicted states to a scalar threat metric Φ^k)
(Note that the superscripts in the Symbol column does not represent a power, but is part of the name, i.e, the first norm, the 2nd norm, the infinity norm. ) Additional norms that may be used include Root Mean Square of predicted state or states over the prediction horizon, and any of the above mentioned norms, with weighting profiles, over the prediction horizon (i.e. state predictions at a chosen time, for
instance, closer to the current vehicle state may be weighted more heavily in the aggregate metric than predicted states at later times or at times other than the specifically chosen time. The chosen time may be other than the current time, such as a time immediately after some other event).
[0091] Just as various vehicle states may be penalized in the objective function without significantly changing the MPC-
generated trajectory plan (as discussed above), these states may also be used somewhat interchangeably to assess threat posed by a given trajectory prediction. Threat assessment based on lateral acceleration (x = y≡ay), front wheel slip
(x = d), and a modified objective function cost ( x = J57 ) have all been compared.
[0092] While lateral acceleration is used in existing threat metrics, front wheel slip may also be used
advantageously in this invention based on three observations. First, front wheel slip is directly tied to, and tends to be a good indicator of, vehicle stability and controllability by front wheel steering. Second, available surface friction places a measureable limit on how large front wheel slip angles can become before loss of control is imminent. This limit provides a useful benchmark against which threat
assessments can be compared to assess maneuver stability (or nearness to instability). Finally, when the cost function's only state objective is to minimize front wheel slip (while remaining within corridor- and actuator-imposed constraints), the path prediction explicitly minimizes the very metric used to assess threat. This hierarchy of objectives — remain within the corridor while minimizing front slip as much as possible — thereby provides a "best case" or minimal-threat assessment from a dynamically-feasible maneuver.
[0093] For some scenarios, however, the controller may not completely satisfy position constraints, making a an
incomplete indicator of the true anticipated threat. These scenarios may arise when complex corridors cause constraints such as maximum input value or maximum input rate to activate. In these situations, the MPC-predicted vehicle path may violate position constraints, making Φa= ^(βpredicted) an
incomplete threat assessment since it does not capture the additional threat posed by the predicted departure from the navigable corridor. To account for such scenarios, an
alternative threat metric similar to the objective function cost may be used, where X = J57, with J57 defined similar to the
MPC objective function. By similar, it is meant of the same form and order. For example, the equation below shows what J51 might look like when the controller objective function
penalizes wheel sideslip a by Rα, steering angle δ by Rg, steering rate Δδ by RΛ6, and constraint violation by ε:
This threat metric, while somewhat more difficult to interpret physically, accounts for the additive presence of the various objective function considerations, such as constraints, input costs, etc, and increases rapidly when constraints are
violated. This rate of intervention is tuned independent of the controller cost function by introducing a modified (and adjustable) constraint violation weight, pSI. The cost-based prediction JSI is related to the predicted front wheel sideslip by
With fa = 0 , this relation allows the cost-based threat
assessment Φ^ to be mapped to an equivalent (and physically- bounded) front-wheel-slip-based assessment Φs via
[0094] Simulation results conducted by the inventors hereof, but not reproduced herein, show how these prediction calculations, threat metrics, and prediction horizons affect the threat assessment. The simulations show that the MPC-based threat assessment provided by the proposed framework gives a similar, albeit more situation- and plant-aware threat
assessment to that based on a constant radius turn in simple (single-obstacle avoidance) scenarios. When generated using the maximum MPC-predicted lateral acceleration (∞-norra) in
these simple scenarios, these predictions provide slightly earlier warning, which explicitly accounts for more complex hazard geometry and more realistic actuator limits. In more complicated/realistic scenarios where multiple hazards are present, threat assessment provided by these inventions has been shown to provide a very accurate assessment of the true threat (or expected nearness to instability) .
[0095] Assessing threat based on a controller-achievable maneuver requires some measure of how well the predicted threat/state represents what the vehicle would actually experience under autonomous control (true threat). This relationship between predicted threat and the controller's ability to maintain true threat at or below this level plays an important role in the semi-autonomous control
implementation discussed in the Semi-Autonomous applications. Threat assessments using various controller and prediction parameters have been compared by the inventors hereof to the true vehicle state and have been shown to provide a reliable estimate of the vehicle's true states under autonomous control .
[0096] When hazard geometry is simple, the threat
assessment generated by methods disclosed herein closely mirrors and slightly precedes (gives more time to prepare for than) a constant radius turn-based assessment. For more complex hazard avoidance scenarios, such as those requiring a lane change maneuver, methods described herein are shown to account for the increased threat, thus providing a significant improvement over simple constant radius turn (CRT) -based assessments .
[0097] Methods disclosed herein have been shown in these simulations to accurately predict an MPC-controlled vehicle's performance as it tracks the MPC-predicted trajectory plan through a constrained corridor. Two metrics have been shown to provide a nearly one-to-one mapping of predicted threat to true threat, suggesting that, if provided full control of a vehicle, this MPC controller can reasonably be expected to
maintain critical vehicle states/threat at or below their predicted values. It is explained below why these inventions' threat assessment may be effectively used for both autonomous and semi-autonomous applications. The inventors hereof have also conducted a comparison of predictions obtained using this objective function setup to objective functions that penalizes lateral acceleration, vehicle roll angle, load transfer, and other states besides front wheel slip (^ >0), with similarly favorable results.
[0098] It is also very beneficial that the threat
assessment may change depending on the circumstances, for example, if a car is traveling through an area corowded with pedestrians. Such a flexibility would be based on a
situational circumstance or severity. A crowded location presents a more severe situation than traveling along an open road. Thus, if there are many obstacles, or if there are several obstacles, but they are moving along complicated paths, the system may note this and invoke a different level of threat assessment, because of the complexity, or severity of the situation.
[0099] One key to these inventions' effectiveness as both a threat assessor and semi-autonomous controller is that it uses the same set of tools, calculations, and model-based
predictions for assessing threat as it does for calculating inputs, and governing human and machine control commands.
Rather than minimizing some arbitrary cost function or value, the trajectory predictor (and actuator command generator in the case of semi-autonomous assistance) minimizes the very characteristics that describe threat. These characteristics, in turn, are based on an accurate model of the vehicle whose performance, stability, and safety may be assessed against an objective, physics-based metric.
[00100] In one particularly useful formulation, the threat assessor combines states, inputs, and constraint violations using a weighting function, that is of the same form and order
as the weighting function used by the trajectory planner and actuator command generator.
EXTENSIONS
[00101] Methods and apparati have been disclosed herein that assess threat in the context of a system with a vehicle moving through an environment. As has been mentioned above, however, these methods and apparati of threat assessment can be used with any system that can be modeled. For instance, they can be used to assess threat in connection with any human — automation system in which the performance of the system being controlled may be reasonably modeled. By performance, it is meant the trajectory of a machine, the motion, velocity or position of a tool, vehicle, etc. The methods and apparati may be used with vehicles of all sorts, including but not limited to terrestrial (ground), aerial, nautical, underwater,
underground, etc. They can be used with heavy equipment, such as agricultural equipment, fork lift trucks, cranes, etc., which not only move from place to place in an environment, but also, or alternatively, change their configuration from one moment to the next, such as when a crane is extending its boom. They may be used with naval vessels. Additionally, they may be used with unmanned or remotely-operated vehicles such as unmanned aerial drones, unmanned ground robots, etc.
[00102] The methods and apparati may be used with surgical machines, which are guided by a human surgeon who manipulates an input device, to drive an output device, such as a scalpel, cauterization tool, stitching machine, etc. The environment of the surgical tool may be considered to include the body being operated upon, instrument supports, etc. They may also be used to control chemical processes, and power plants, such as nuclear, electric, etc.
[00103] The inventions disclosed herein are also useful with respect to devices that do not move from one point to another in an environment, and are also useful in connection with
large systems, such as power plants, and also processes, such as chemical processes.
[00104] The terms: states, control inputs, operator inputs, trajectory, and regions, are all general to modeled systems and therefore translate directly to surgical devices, chemical processes, etc., as used herein and in the accompanying claims .
[00105] A region of the state space bounded by state
constraints is also generic to modeled systems. Corridor, as used herein, is a convenient way of visualizing a two- dimensional region and may therefore not be so intuitively useful in other applications. However, it may be meant to use a bounded two-dimensional corridor in a state space that is different from a traditional, familiar two-dimensional
physical space, such as that in which a vehicle travels. Thus, as used in the claims, it means a bounded two-dimensional space, analogous to a roadway in terrestrial two-dimensional space.
[00106] A system that does not move within an environment still needs to account for the influence of external processes and disturbances upon it. Although these are not obstacles per se, they still represent threats to system success. In a chemical process, such an obstacle-like threat could be an external agency that dumps chemicals at different intervals into the mixture under control, or an uncontrolled temperature variation in the environment of the plant. Regarding a
chemical process, the term process plant is used herein and in the claims to refer to the volume of material that is being transformed and manipulated according to a chemical process, which may be in a stationary vessel, or moving along in a series of pipes and vessels, over time. The environment in which the process plant operates would include the vessels and atmospheres in which the process plant exists. Environmental disturbances, or hazards, could include ambient temperature influences, chemical pollutants, vibrations of equipment, etc. Rather than the locations of obstacles in physical space being
relevant to the control of a process, inventions disclosed herein use the coordinates of such influences, in whatever coordinate system is relevant for understanding and evaluating their effect upon the process under control. For instance, the coordinates of chemical process hazards may relate to which vessel, or run of conduit a disturbance affects.
[00107] With respect to other systems, such as power plants, analogous factors will be understood by those skilled in the art.
[00108] As has been mentioned, use of a threat assessment that is closely tied to the means by which an optimal
trajectory is established enables an optimal framework of assisting a device operator. This framework explicitly
considers human and machine dynamics without presuming
operator intentions or limiting the avoidance maneuver (and its associated threat assessment) to a specific path. It provides a unified framework that allows for various modes and levels of operator assistance, from operator warning to stability control to passive intervention, to active semi- autonomous control, and finally, to autonomous machine
operation. Autonomous operation may be local or remote.
[00109] Remotely operated systems typically exhibit a time lag. Because this method for threat assessment and semi- autonomous operator assistance is based on a model predictive control prediction, it is particularly well suited for
unmanned and remotely or ' teleoperated' systems such as
Unmanned Ground Vehicles (UGV s), Unmanned Aerial Vehicles (UAV s), Unmanned Underwater Vehicles (UUVs), or remote surgical equipment. In relaying signals between the operator and the vehicle, such systems frequently experience
communication latencies and time delays. These latencies and time delays can cause problems for prior art systems seeking to assist the human driver. Such systems often require
additional modules to mitigate the effects of latency and time delays. In contrast, the framework described here may be configured to explicitly incorporate these time delays into
the optimal trajectory prediction, thereby eliminating the need for additional modules.
Operator Assistance
[00110] Known systems may be illustrated with reference to Fig. 10. Known systems use distinct and sometimes competing modules to assist an operator. Such distinct, independent systems 1000 can include, but are not limited to: warning devices 1002, a yaw stability controller 1004; a roll
stability controller 1006, a traction controller 1008, a cruise controller 1010 and a lane-assist controller 1012. All of which may work independently, and their combined output, (some of which are inputs to actuators that control the vehicle) can lead to inconsistent, unpredictable, unintended or suboptimal outcome 1014. For instance, the roll stability controller may output a steering command that is inconsistent with a steering command generated by the lane assist
controller.
[00111] In contrast, as shown in Fig. 11, the inventions described here may operate as a unitary, consistent whole, 1100, in conjunction with, or independent of, the human operator.
[00112] The MPC constitutes a state trajectory and control input planner 1102, which generates 1102 an optimal trajectory for the device, from current position through a time horizon, together with the device control inputs necessary to follow that trajectory.
[00113] In a passive mode, these inventions constitute a physically-accurate, predictive and flexible means of
predicting the threat posed to the vehicle given the driver's current and past performance, current and past vehicle state evolution, and environmental features/constraints. It is physically accurate because it is based on a forward-simulated vehicle model. It is predictive, because the vehicle model is simulated over a future time horizon. It is flexible because
threat and intervention decisions are based on a corridor — rather than path characteristics.
[00114] Warning and operator feedback systems 1104 may inform any of a number of operator warning devices, including audible, visual, haptic, olfactory, etc.
[00115] In addition to providing a predictive, flexible, and accurate passive warning mechanism 1104, the inventions may be configured to predictively and semi-autonomousIy act as a device stability controller 1106. Examples of stability controllers that this system might replace include, but are not limited to yaw stability controllers, roll stability controllers, anti-lock brakes and traction controls. Using various mutually compatible actuation modes, including, for example, differential braking and steering adjustments, this framework predicts vehicle state evolution over a projected time horizon and adjusts the current vehicle state to keep both current and predicted stability-critical states (such as yaw angle, wheel slip angles, etc.) below critical levels.
[00116] Another related intervention mode that these
inventions are capable of is passive driver assistance 1108. These subsystems can apply various degrees of resistance, such as to steering wheel resistance torque overlays, braking (anti-lock brakes) traction control, yaw stability control, acceleration inhibitors, to discourage the operator from further increasing threat. (Note that there is some overlap between passive and reactive assistance systems and stability controllers . )
[00117] Active intervention modes 1110 go beyond providing resistive feedback and actively initiate steering and/or braking commands .
[00118] The inventions are well-suited to semi-autonomousIy assist the driver; utilizing sensory information related to the vehicle surroundings and a prediction of a safe vehicle trajectory through those surroundings to exert appropriate actuator effort and avoid impending accidents. Sensory
information would include data related to nearby vehicles, pedestrians, road edges, and other salient features to assess accident threat. A second key advantage is that the invention may be configured to operate only during instances of
significant threat: giving the driver full control of the vehicle in low threat situations but applying appropriate levels of computer-controlled actuator effort during high threat situations.
[00119] The corridor-based threat assessment upon which this intervention is based presents a key advantage over existing semi-autonomous safety systems in that using a corridor (as opposed to a path) allows intervention that does not
unnecessarily constrain the human operator to a specific (and rather arbitrary) path.
[00120] As a semi-autonomous controller, the inventions allow for human-controlled, computer-controlled, and shared human/computer vehicle operation. The form of the
intervention law modulating control system authority
allocation is configurable and can differ for different actuators (i.e. a vehicle with both active steering and braking can have distinct intervention laws defined for the steering actuator and the braking actuators ) . The
intervention law can also be defined to adapt to driver performance based on an assessment of driver skill, and/or to include considerations for driver preference, environmental conditions, previous threat metric values, previous control inputs, and other factors. The intervention law can also take on dynamics of its own, exhibiting, for example, hysteretic and higher-order behavior as a function of past, current, and predictive threat. See Fig. 1.
[00121] Finally, the inventions are capable of fully- autonomous vehicle control. That is, in high-threat scenarios or when desired by the human operator, this system can take full control of the vehicle, safely navigating it through the environment while avoiding both collisions and loss of control .
[00122] This multiplicity of highly competent options enables a unitary, integrated, mutually consistent, optimal operator assistance system. Because the modules for
effectuating each degree of autonomy in operation described above are each based on the same optimal control sequence of states and inputs, and are all tied together by virtue of the threat assessment apparatus and method that is also based on the same sequence of states and inputs, the optimal set of inputs is optimal and internally consistent. The passive assistance portions of the system will not generate
conflicting actuator inputs with other passive assistance portions, or with the active assistance portions. Thus, the system can allow and direct control to flow seamlessly from a mere warning, to passive assistance, stability control, active assistance, semi-autonomous and fully autonomous, and back again, over and over again, automatically. All of the threat response activities are mutually consistent. This permits a truly, reliably user configurable semi-autonomous system. All of this flexibility flows from the use of generating an appropriate and active control input based on predicted threat.
[00123] This unification through the optimal trajectory of position, control and state sequences is highly significant. It means that one system can perform equally well the duties currently assigned to many disparate systems. For example, a vehicle that was previously equipped with separate driver warning, traction control, and yaw stability control systems (whose assistance can often conflict), may, with the present inventions, perform each of these functions (and more still) with a single, unified, and in some sense optimal framework. Thus, that a single system reliably, elegantly, and highly satisfactorily performs any of these tasks individually, and all of the tasks together, which otherwise would require multiple systems is significant. Such an invention is far more than simply the sum of its parts.
[00124] It will be understood that the MPC path and
trajectory planner 1102 and the threat assessor 1103 and
actuator controllers 1104, 1106, 1108, 1110 may be microprocessor based such as a computer having a central processing unit, memory (RAM and/or ROM), and associated input and output buses. They may be application specific integrated circuits or may be formed of any other logic devices known in the art. They may be a portion of a central vehicle main control unit, an interactive vehicle dynamics module, a main safety controller, or may be stand=-alone controllers. They may each have associated memories, may share a central memory or some combination of both. These controllers may perform various different sensing system operations. The operations may be performed sequentially or simultaneously. The
controller and threat assessor may have a driver input.
[00125] The foregoing has described the computer systems that conduct method steps discussed herein relatively
generically. It will be understood by those skilled in the art that each of the processes described herein can be performed by a dedicated specialized processor, or by a properly
programmed general purpose digital computer, or by some computing device that is more or less complex and specialized than either of the foregoing. As such, where in the claims it is stated that a processor performs a specific function, it will be understood that such a processor may perform only the function stated therein, or additional, and even all processes mentioned within the claim. Assignment of such processes to one or more processing devices is routine and would be
conducted by a skilled person according to processing speed, weight.
[00126] This disclosure describes and discloses more than one invention. The inventions are set forth in the claims of this and related documents, not only as filed, but also as developed during prosecution of any patent application based on this disclosure. The inventors intend to claim all of the various inventions to the limits permitted by the prior art, as it is subsequently determined to be. No feature described herein is essential to each invention disclosed herein. Thus, the inventors intend that no features described herein, but
not claimed in any particular claim of any patent based on this disclosure, should be incorporated into any such claim.
[00127] Some assemblies of hardware, or groups of steps, are referred to herein as an invention. However, this is not an admission that any such assemblies or groups are necessarily patentably distinct inventions, particularly as contemplated by laws and regulations regarding the number of inventions that will be examined in one patent application, or unity of invention. It is intended to be a short way of saying an embodiment of an invention.
[00128] An abstract is submitted herewith. It is emphasized that this abstract is being provided to comply with the rule requiring an abstract that will allow examiners and other searchers to quickly ascertain the subject matter of the technical disclosure. It is submitted with the understanding that it will not be used to interpret or limit the scope or meaning of the claims, as promised by the Patent Office's rule.
[00129] The foregoing discussion should be understood as illustrative and should not be considered to be limiting in any sense. While the inventions have been particularly shown and described with references to preferred embodiments thereof, it will be understood by those skilled in the art that various changes in form and details may be made therein without departing from the spirit and scope of the inventions as defined by the claims.
[00130] The corresponding structures, materials, acts and equivalents of all means or step plus function elements in the claims below are intended to include any structure, material, or acts for performing the functions in combination with other claimed elements as specifically claimed.
[00131] Having described the invention, what is claimed is:
Claims
1. A method for generating a threat response, related to threat posed to a mechanical device, comprising the steps of: a. using a model of the device, and an objective function, generating an optimal sequence of predicted device controller inputs, generated so as to optimize predicted performance of the model over a time horizon over which the device model is simulated and satisfy device dynamic constraints, device input constraints, and environmental constraints; b. generating an optimal sequence of predicted device states resulting from applying the sequence of optimal device inputs to a model of the device; c. combining at least one of: the predicted states and the predicted inputs, into a scalar threat
assessment metric; and, d. using the scalar threat assessment metric determining a type and degree of operator assistance to provide as the threat response and executing the
operator assistance.
2. The method of claim 1, the step of determining comprising choosing an operator assistance from a set of at least two mutually consistent threat responses, each response of the set being based on the same threat assessment metric.
3. The method of claim 1, the threat response being at least one selected from the group of mutually consistent responses consisting of: i. triggering a warning for a human device operator; ii. triggering a collision prevention
mechanism; iii. triggering a collision mitigation
mechanism; iv. regulating an input of a human device operator; v. modifying an input of a human device
operator; vi. regulating an input of the optimal
controller; vii. modifying an input of the optimal
controller; and viii. triggering autonomous control of the device by the optimal controller.
4. The method of claim 3, the step of combining at least one of: the predicted states and the predicted inputs, into a scalar threat assessment metric comprising; determining a threat cost at each prediction step in the optimal sequence using a weighting function of the same form as the objective function that was used to generate a sequence of at least one optimal device state and optimal device inputs and combining the sequence of determined threat costs to generate an
aggregate threat assessment metric.
5. The method of claim 4, the step of generating the aggregate threat assessment metric comprising basing the generation on at least one norm, selected from the group consisting of: a. a maximum predicted value of at least one threat cost over a prediction horizon; b. an average predicted value of at least one threat cost over a prediction horizon; and c. a Root Mean Square (RMS) of predicted value of at least one threat cost over a prediction horizon.
6. The method of claim 4, the step of generating the aggregate threat assessment metric comprising basing the generation on at least one norm, selected from the group consisting of: a. a maximum predicted value of at least one state over a prediction horizon; b. an average predicted value of at least one state over a prediction horizon c. a Root Mean Square (RMS) of predicted value of at least one state over a prediction horizon; d. a maximum predicted value of at least one input over a prediction horizon; e. an average predicted value of at least one input over a prediction horizon f. a Root Mean Square (RMS) of predicted value of at least one input over a prediction horizon; and g. a maximum predicted value of at least one input and at least one state over a prediction horizon; h. an average predicted value of at least one input and at least one state over a prediction horizon i. a Root Mean Square (RMS) of predicted value of at least one input and at least one state input over a prediction horizon; and j . a weighted prediction of any of the factors of states and inputs listed in sub-paragraphs a-i of this claim, formed by weighting predictions that are at a chosen time more heavily in the aggregate threat metric than a predicted factor at a different times.
7. The method of claim 3, the warning for a human device operator comprising a signal selected from the group
consisting of: an audible signal, a visual signal, a haptic signal, and an olfactory signal.
8. The method of claim 3, the humanly perceptible signal varying in at least one of frequency and amplitude based on the threat assessment metric.
9. The method of claim 3, the control assistance
comprising a passive device control assistance.
10. The method of claim 3, the control assistance comprising a reactive device control assistance.
11. The method of claim 3, the control assistance comprising an active device control assistance.
12. The method of claim 10, the device comprising a ground vehicle, the active device control assistance
comprising at least one selected from the group consisting of: traction control, adaptive cruise control yaw stability control, roll stability control, electronic stability control; lane keeping and obstacle avoidance.
13. The method of claim 1, the device comprising a ground vehicle, the operator assistance comprising at least one selected from the group consisting of: resistance torques on a steering input member; resistance torques on a braking input member; resistance torques on an acceleration input member; acceleration torques; braking torques; priming wheel brakes, preparing safety devices and modifying device properties.
14. The method of claim 1, the operator assistance comprising an active device modulation.
15. The method of claim 14, the device comprising a ground vehicle, the device modulation signal comprising suspension modification.
16. The method of claim I1 the device comprising a ground vehicle.
17. The method of claim 1, the device comprising an aerial vehicle.
18. The method of claim I1 the device comprising a naval vessel.
19. The method of claim 1, the device comprising heavy construction equipment.
20. The method of claim I1 the device comprising a fork lift truck.
20. The method of claim 1, the device comprising a surgical device.
21. The method of claim I1 the device comprising a teleoperated device residing in the environment, which device environment is distant from an environment in which a human device operator resides.
Priority Applications (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US13/254,761 US8437890B2 (en) | 2009-03-05 | 2010-07-15 | Integrated framework for vehicle operator assistance based on a trajectory prediction and threat assessment |
US13/859,203 US8744648B2 (en) | 2009-03-05 | 2013-04-09 | Integrated framework for vehicle operator assistance based on a trajectory prediction and threat assessment |
Applications Claiming Priority (6)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
US27093309P | 2009-07-15 | 2009-07-15 | |
US61/270,933 | 2009-07-15 | ||
USPCT/US2010/0025206 | 2010-02-24 | ||
US12/711,935 | 2010-02-24 | ||
PCT/US2010/025206 WO2010101749A1 (en) | 2009-03-05 | 2010-02-24 | Predictive semi-autonomous vehicle navigation system |
US12/711,935 US20100228427A1 (en) | 2009-03-05 | 2010-02-24 | Predictive semi-autonomous vehicle navigation system |
Related Parent Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US12/711,935 Continuation-In-Part US20100228427A1 (en) | 2009-03-05 | 2010-02-24 | Predictive semi-autonomous vehicle navigation system |
Related Child Applications (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
US13/254,761 A-371-Of-International US8437890B2 (en) | 2009-03-05 | 2010-07-15 | Integrated framework for vehicle operator assistance based on a trajectory prediction and threat assessment |
US13/859,203 Continuation US8744648B2 (en) | 2009-03-05 | 2013-04-09 | Integrated framework for vehicle operator assistance based on a trajectory prediction and threat assessment |
Publications (1)
Publication Number | Publication Date |
---|---|
WO2011009011A1 true WO2011009011A1 (en) | 2011-01-20 |
Family
ID=43449811
Family Applications (2)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2010/042203 WO2011009011A1 (en) | 2009-03-05 | 2010-07-15 | An integrated framework for vehicle operator assistance based on a trajectory prediction and threat assessment |
PCT/US2010/042201 WO2011009009A1 (en) | 2009-03-05 | 2010-07-15 | Methods and apparati for predicting and quantifying threat being experienced by a modeled system |
Family Applications After (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
PCT/US2010/042201 WO2011009009A1 (en) | 2009-03-05 | 2010-07-15 | Methods and apparati for predicting and quantifying threat being experienced by a modeled system |
Country Status (1)
Country | Link |
---|---|
WO (2) | WO2011009011A1 (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8744648B2 (en) | 2009-03-05 | 2014-06-03 | Massachusetts Institute Of Technology | Integrated framework for vehicle operator assistance based on a trajectory prediction and threat assessment |
US9342989B2 (en) | 2012-06-29 | 2016-05-17 | Bae Systems Information And Electronic Systems Integration Inc. | Radio-enabled collision avoidance system |
CN110244721A (en) * | 2019-06-05 | 2019-09-17 | 杭州飞步科技有限公司 | Automatic Pilot control method, device, equipment and storage medium |
CN110442908A (en) * | 2019-07-02 | 2019-11-12 | 合肥工业大学 | For assessing the method, system and storage medium of aircraft track risk |
CN112148008A (en) * | 2020-09-18 | 2020-12-29 | 中国航空无线电电子研究所 | Real-time unmanned aerial vehicle path prediction method based on deep reinforcement learning |
CN116680542A (en) * | 2023-06-27 | 2023-09-01 | 北京五木恒润科技有限公司 | Method and system for generating trunk branch situation and bypass branch situation |
Families Citing this family (17)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP2615598B1 (en) | 2012-01-11 | 2017-12-06 | Honda Research Institute Europe GmbH | Vehicle with computing means for monitoring and predicting traffic participant objects |
US9031892B2 (en) | 2012-04-19 | 2015-05-12 | Invensys Systems, Inc. | Real time safety management system and method |
CN102768539B (en) * | 2012-06-26 | 2014-08-06 | 哈尔滨工程大学 | AUV (autonomous underwater vehicle) three-dimension curve path tracking control method based on iteration |
US8898016B2 (en) | 2013-03-15 | 2014-11-25 | Applied Minds, Llc | Method and apparatus for two-stage planning |
US11022446B2 (en) | 2013-03-15 | 2021-06-01 | Applied Invention, Llc | Method and apparatus for two-stage planning |
SE537674C2 (en) * | 2013-03-19 | 2015-09-29 | Scania Cv Ab | Control system for autonomous vehicles, and a method for the control system |
EP2848487B1 (en) * | 2013-09-12 | 2020-03-18 | Volvo Car Corporation | Manoeuvre generation for automated drive |
US9666069B2 (en) * | 2014-02-14 | 2017-05-30 | Ford Global Technologies, Llc | Autonomous vehicle handling and performance adjustment |
EP3001272B1 (en) * | 2014-09-26 | 2017-04-12 | Volvo Car Corporation | Method of trajectory planning for yielding manoeuvres |
FR3063265B1 (en) * | 2017-02-28 | 2019-04-05 | Renault S.A.S | DEVICE FOR CONTROLLING THE TRACK OF A VEHICLE |
CN107103152B (en) * | 2017-05-16 | 2021-02-02 | 上海航天控制技术研究所 | Thruster installation layout method beneficial to inhibiting liquid from shaking |
CN107271969B (en) * | 2017-07-17 | 2019-05-21 | 电子科技大学 | Interference source configuration method under safe corridor task object |
DE102018109883A1 (en) | 2018-04-24 | 2018-12-20 | Continental Teves Ag & Co. Ohg | Method and device for the cooperative tuning of future driving maneuvers of a vehicle with foreign maneuvers of at least one other vehicle |
CN109360445B (en) * | 2018-07-09 | 2021-02-09 | 重庆大学 | High-speed lane change risk detection method based on transverse and longitudinal kinematic feature distribution |
KR102505300B1 (en) * | 2018-12-18 | 2023-03-02 | 모셔널 에이디 엘엘씨 | Behavior of vehicles using multiple motion constraints |
US11899464B2 (en) | 2018-12-18 | 2024-02-13 | Motional Ad Llc | Operation of a vehicle using motion planning with machine learning |
CN109445463B (en) * | 2018-12-21 | 2021-07-27 | 山东理工大学 | Unmanned aerial vehicle dynamic route planning method |
Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020022927A1 (en) * | 1993-08-11 | 2002-02-21 | Lemelson Jerome H. | GPS vehicle collision avoidance warning and control system and method |
US20030109780A1 (en) * | 2001-06-07 | 2003-06-12 | Inria Roquencourt | Methods and apparatus for surgical planning |
US20070112476A1 (en) * | 1997-08-01 | 2007-05-17 | American Calcar Inc. | Centralized control and management system for automobiles |
US20080208409A1 (en) * | 2003-02-10 | 2008-08-28 | Nissan Motor Co., Ltd. | Vehicle dynamics control apparatus |
US7512487B1 (en) * | 2006-11-02 | 2009-03-31 | Google Inc. | Adaptive and personalized navigation system |
US20090174540A1 (en) * | 2008-01-04 | 2009-07-09 | Smith Alexander E | Method and apparatus to determine vehicle intent |
Family Cites Families (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US6216053B1 (en) * | 1992-11-09 | 2001-04-10 | Lextron, Inc. | Apparatus and method for uniformly delivering feed rations along a feedbunk using global positioning system |
US7630806B2 (en) * | 1994-05-23 | 2009-12-08 | Automotive Technologies International, Inc. | System and method for detecting and protecting pedestrians |
US6173215B1 (en) * | 1997-12-19 | 2001-01-09 | Caterpillar Inc. | Method for determining a desired response to detection of an obstacle |
US20030229552A1 (en) * | 2002-06-05 | 2003-12-11 | Lebaric Katarina J. | System and method for deal-making decision optimization |
US20070247517A1 (en) * | 2004-08-23 | 2007-10-25 | Sarnoff Corporation | Method and apparatus for producing a fused image |
US7248952B2 (en) * | 2005-02-17 | 2007-07-24 | Northrop Grumman Corporation | Mixed integer linear programming trajectory generation for autonomous nap-of-the-earth flight in a threat environment |
-
2010
- 2010-07-15 WO PCT/US2010/042203 patent/WO2011009011A1/en active Application Filing
- 2010-07-15 WO PCT/US2010/042201 patent/WO2011009009A1/en active Application Filing
Patent Citations (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US20020022927A1 (en) * | 1993-08-11 | 2002-02-21 | Lemelson Jerome H. | GPS vehicle collision avoidance warning and control system and method |
US20070112476A1 (en) * | 1997-08-01 | 2007-05-17 | American Calcar Inc. | Centralized control and management system for automobiles |
US20030109780A1 (en) * | 2001-06-07 | 2003-06-12 | Inria Roquencourt | Methods and apparatus for surgical planning |
US20080208409A1 (en) * | 2003-02-10 | 2008-08-28 | Nissan Motor Co., Ltd. | Vehicle dynamics control apparatus |
US7512487B1 (en) * | 2006-11-02 | 2009-03-31 | Google Inc. | Adaptive and personalized navigation system |
US20090174540A1 (en) * | 2008-01-04 | 2009-07-09 | Smith Alexander E | Method and apparatus to determine vehicle intent |
Cited By (9)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
US8744648B2 (en) | 2009-03-05 | 2014-06-03 | Massachusetts Institute Of Technology | Integrated framework for vehicle operator assistance based on a trajectory prediction and threat assessment |
US9342989B2 (en) | 2012-06-29 | 2016-05-17 | Bae Systems Information And Electronic Systems Integration Inc. | Radio-enabled collision avoidance system |
CN110244721A (en) * | 2019-06-05 | 2019-09-17 | 杭州飞步科技有限公司 | Automatic Pilot control method, device, equipment and storage medium |
CN110244721B (en) * | 2019-06-05 | 2022-04-12 | 杭州飞步科技有限公司 | Automatic driving control method, device, equipment and storage medium |
CN110442908A (en) * | 2019-07-02 | 2019-11-12 | 合肥工业大学 | For assessing the method, system and storage medium of aircraft track risk |
CN110442908B (en) * | 2019-07-02 | 2022-09-27 | 合肥工业大学 | Method, system and storage medium for assessing risk of trajectory of aircraft |
CN112148008A (en) * | 2020-09-18 | 2020-12-29 | 中国航空无线电电子研究所 | Real-time unmanned aerial vehicle path prediction method based on deep reinforcement learning |
CN116680542A (en) * | 2023-06-27 | 2023-09-01 | 北京五木恒润科技有限公司 | Method and system for generating trunk branch situation and bypass branch situation |
CN116680542B (en) * | 2023-06-27 | 2024-01-09 | 北京五木恒润科技有限公司 | Method and system for generating trunk branch situation and bypass branch situation |
Also Published As
Publication number | Publication date |
---|---|
WO2011009009A1 (en) | 2011-01-20 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
US8543261B2 (en) | Methods and apparati for predicting and quantifying threat being experienced by a modeled system | |
WO2011009011A1 (en) | An integrated framework for vehicle operator assistance based on a trajectory prediction and threat assessment | |
Carvalho et al. | Automated driving: The role of forecasts and uncertainty—A control perspective | |
Rasekhipour et al. | A potential field-based model predictive path-planning controller for autonomous road vehicles | |
Anderson et al. | An optimal-control-based framework for trajectory planning, threat assessment, and semi-autonomous control of passenger vehicles in hazard avoidance scenarios | |
Llorca et al. | Autonomous pedestrian collision avoidance using a fuzzy steering controller | |
Anderson et al. | The intelligent copilot: A constraint-based approach to shared-adaptive control of ground vehicles | |
Anderson et al. | Constraint-based planning and control for safe, semi-autonomous operation of vehicles | |
Lima et al. | Spatial model predictive control for smooth and accurate steering of an autonomous truck | |
Wang et al. | Modeling and field experiments on autonomous vehicle lane changing with surrounding human‐driven vehicles | |
RU2721860C2 (en) | Steering column torque control system and method | |
Hosseini et al. | Predictive haptic feedback for safe lateral control of teleoperated road vehicles in urban areas | |
US20200216090A1 (en) | Systems and methods for shared control of a vehicle | |
Berntorp et al. | Positive invariant sets for safe integrated vehicle motion planning and control | |
CN114761895A (en) | Direct and indirect control of hybrid automated fleet | |
Morales et al. | Proactive driving modeling in blind intersections based on expert driver data | |
Micheli et al. | NMPC trajectory planner for urban autonomous driving | |
Elmi et al. | Path planning using model predictive controller based on potential field for autonomous vehicles | |
Xu et al. | Autonomous vehicle active safety system based on path planning and predictive control | |
CN113200054A (en) | Path planning method and system for automatic driving takeover | |
Anderson et al. | A unified approach to semi-autonomous control of passenger vehicles in hazard avoidance scenarios | |
Xue et al. | An adaptive model predictive approach for automated vehicle control in fallback procedure based on virtual vehicle scheme | |
Anderson et al. | Semi-autonomous stability control and hazard avoidance for manned and unmanned ground vehicles | |
Nietzschmann et al. | Trajectory optimization for car-like vehicles in structured and semi-structured environments | |
Rasekhipour | Prioritized Obstacle Avoidance in Motion Planning of Autonomous Vehicles |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 10800581 Country of ref document: EP Kind code of ref document: A1 |
|
WWE | Wipo information: entry into national phase |
Ref document number: 13254761 Country of ref document: US |
|
122 | Ep: pct application non-entry in european phase |
Ref document number: 10800581 Country of ref document: EP Kind code of ref document: A1 |
|
NENP | Non-entry into the national phase |
Ref country code: DE |