CN116767272A - Adaptation of driving behavior of autonomous vehicles - Google Patents
Adaptation of driving behavior of autonomous vehicles Download PDFInfo
- Publication number
- CN116767272A CN116767272A CN202310196851.7A CN202310196851A CN116767272A CN 116767272 A CN116767272 A CN 116767272A CN 202310196851 A CN202310196851 A CN 202310196851A CN 116767272 A CN116767272 A CN 116767272A
- Authority
- CN
- China
- Prior art keywords
- autonomous vehicle
- driving
- vehicle
- traffic
- intersection
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 230000006978 adaptation Effects 0.000 title description 2
- 238000000034 method Methods 0.000 claims abstract description 35
- 230000006399 behavior Effects 0.000 claims description 31
- 230000001934 delay Effects 0.000 claims description 15
- 238000011156 evaluation Methods 0.000 claims description 12
- 238000005457 optimization Methods 0.000 claims description 12
- 238000012549 training Methods 0.000 claims description 11
- 238000013528 artificial neural network Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 4
- 230000002787 reinforcement Effects 0.000 claims description 3
- 238000004891 communication Methods 0.000 description 11
- 238000010801 machine learning Methods 0.000 description 3
- 238000002604 ultrasonography Methods 0.000 description 3
- 238000013459 approach Methods 0.000 description 1
- 238000013473 artificial intelligence Methods 0.000 description 1
- 238000001514 detection method Methods 0.000 description 1
- 238000003909 pattern recognition Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
Classifications
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W60/00—Drive control systems specially adapted for autonomous road vehicles
- B60W60/001—Planning or execution of driving tasks
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W30/00—Purposes of road vehicle drive control systems not related to the control of a particular sub-unit, e.g. of systems using conjoint control of vehicle sub-units
- B60W30/18—Propelling the vehicle
- B60W30/18009—Propelling the vehicle related to particular drive situations
- B60W30/18154—Approaching an intersection
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W30/00—Purposes of road vehicle drive control systems not related to the control of a particular sub-unit, e.g. of systems using conjoint control of vehicle sub-units
- B60W30/18—Propelling the vehicle
- B60W30/18009—Propelling the vehicle related to particular drive situations
- B60W30/18159—Traversing an intersection
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W50/00—Details of control systems for road vehicle drive control not related to the control of a particular sub-unit, e.g. process diagnostic or vehicle driver interfaces
- B60W2050/0062—Adapting control system settings
- B60W2050/0075—Automatic parameter input, automatic initialising or calibrating means
- B60W2050/0083—Setting, resetting, calibration
- B60W2050/0088—Adaptive recalibration
-
- B—PERFORMING OPERATIONS; TRANSPORTING
- B60—VEHICLES IN GENERAL
- B60W—CONJOINT CONTROL OF VEHICLE SUB-UNITS OF DIFFERENT TYPE OR DIFFERENT FUNCTION; CONTROL SYSTEMS SPECIALLY ADAPTED FOR HYBRID VEHICLES; ROAD VEHICLE DRIVE CONTROL SYSTEMS FOR PURPOSES NOT RELATED TO THE CONTROL OF A PARTICULAR SUB-UNIT
- B60W2554/00—Input parameters relating to objects
- B60W2554/40—Dynamic objects, e.g. animals, windblown objects
- B60W2554/408—Traffic behavior, e.g. swarm
Landscapes
- Engineering & Computer Science (AREA)
- Automation & Control Theory (AREA)
- Transportation (AREA)
- Mechanical Engineering (AREA)
- Human Computer Interaction (AREA)
- Traffic Control Systems (AREA)
Abstract
The invention relates to a method for adapting the driving behavior of an autonomous vehicle (217). In the method, a traffic condition is defined that includes a driving maneuver to be performed by an autonomous vehicle (217). Furthermore, a model for performing a driving maneuver under the traffic conditions is trained. The model considers and evaluates traffic flows associated with the traffic conditions. The execution of the driving maneuver is trained in a manner that optimizes the traffic flow based on its assessment. In the traffic situation, a driving maneuver is performed by the autonomous vehicle (217) depending on the traffic flow according to the model.
Description
Technical Field
The present invention relates to a method of adapting the driving behaviour of an autonomous vehicle.
Background
An autonomous vehicle is herein referred to as an autonomous vehicle, i.e. a vehicle that independently performs a driving maneuver without driver intervention. Independent execution of travel maneuvers places various demands on autonomous vehicles. In particular, traffic conditions, including other road users and their movements, must be detected and evaluated. For this purpose, autonomous vehicles have a variety of sensors to detect their surroundings, such as cameras, lidar sensors, radar sensors and/or ultrasound sensors. The sensor signals of these sensors are evaluated and judged by the autonomous vehicle to perform travel maneuvers that are tailored to traffic conditions. The driving maneuver performed by the autonomous vehicle is particularly important here without jeopardizing other road users. As a result, autonomous vehicles often have a so-called defensive driving style, which places safety problems in the first place. However, excessive defensive driving style can lead to traffic obstruction. In addition, excessive defensive driving style can annoy other road users and entice them to take dangerous driving maneuvers, such as dangerous overtaking maneuvers.
EP 3598414A1 discloses a method of predicting the trajectory of at least one road user in order to avoid collision of a subject vehicle, in particular an autonomous vehicle, with the road user. Here, a group of possible trajectories of the road user is determined and at least one trajectory of the road user is predicted therefrom. At least one action is performed by the subject vehicle based on the at least one predicted trajectory.
WO 2020/040975 A1 discloses a method in which data detected by vehicle sensors are stored, which data describe the driving behaviour of a manually controlled vehicle in traffic conditions and are used, for example, to train a machine learning model for controlling an autonomous vehicle or to support the driver of the vehicle. In order to limit the amount of data to be stored, high resolution data of the driving behavior is stored only when the driving behavior has a special meaning, for example when the driving behavior deviates significantly from the predicted driving behavior.
Disclosure of Invention
The present invention is based on the object of reducing traffic impediments caused by the driving behaviour of autonomous vehicles.
According to the invention, this object is achieved by a method for adapting the driving behavior of an autonomous vehicle, wherein
Defining a traffic condition including a driving maneuver to be performed by the autonomous vehicle,
training a model for performing driving maneuvers in the traffic situation,
wherein the model considers and evaluates traffic flows associated with the traffic situation and trains the execution of driving maneuvers in a manner that optimizes traffic flows according to its evaluation,
and in the traffic situation, performing, by the autonomous vehicle, a driving maneuver depending on the traffic flow according to the model.
According to the method of the invention, the driving behavior of the autonomous vehicle under the traffic condition is adjusted in dependence on the influence of the driving behavior of the autonomous vehicle on the traffic flow associated with the traffic condition. In this case, a model of the driving behavior of the autonomous vehicle under traffic conditions is trained for the purpose of optimizing the traffic flow. This distinguishes the method according to the invention from the prior art known from e.g. EP 3598414A1, which is intended to avoid collisions of the subject vehicle, in particular an autonomous vehicle, with other road users.
In training the model, the influence of the driving behavior of the autonomous vehicle on the traffic flow is evaluated and optimized according to appropriate optimization parameters. Thus, the driving behavior of an autonomous vehicle and its impact during model training are retrospectively considered and evaluated in terms of optimization of traffic flow. This retrospective consideration and evaluation of the driving behavior is a further distinction from the prior art known, for example, from EP 3598414A1 and WO 2020/040975 A1.
In one embodiment of the invention, the model is based on an artificial neural network. Artificial neural networks are artificial intelligence tools, for example, for pattern recognition and machine learning. They are therefore also particularly suitable for machine learning of driving behavior on the basis of training data.
In another embodiment of the invention, the model is trained with data recorded in reality. In particular, the model can here in particular reward the execution of driving maneuvers that have been executed in reality by non-autonomous vehicles by reinforcement learning (english Reinforced Learning).
The above-described embodiments of the present invention make it possible for the driving behavior of an autonomous vehicle to be adapted to the driving behavior of a non-autonomous vehicle. In particular, an excessively defensive driving behavior of the autonomous vehicle, which may lead to unnecessary travel delays of the autonomous vehicle and travel obstructions of other vehicles, can thereby be avoided.
In another embodiment of the invention, the traffic flow in the traffic condition is detected with at least one sensor of the autonomous vehicle. Sensors that can be considered for detecting traffic flow are, for example, cameras, lidar sensors, radar sensors and ultrasound sensors. Autonomous vehicles typically have a plurality of such sensors.
In another embodiment of the invention, both the travel delays of the autonomous vehicle and the travel delays of the other vehicles caused by the driving behavior of the autonomous vehicle are taken into account when evaluating the traffic flow. For example, pareto optimization is performed in optimizing traffic flow, with the objective of minimizing the travel delays of the autonomous vehicle and the travel delays of other vehicles caused by the driving behavior of the autonomous vehicle.
Pareto optimization, also known as multi-objective optimization, refers to solving an optimization problem with multiple objectives that often conflict with each other. In this case, these objectives are the minimization of the travel delays of different vehicles, including the autonomous vehicle itself. According to the above-described embodiments of the present invention, the travel delay of the autonomous vehicle itself is also incorporated into the evaluation when evaluating and optimizing the traffic flow. Excessive defensive driving behavior of the autonomous vehicle can thereby also be avoided. For example, to avoid excessive travel delays of the autonomous vehicle, slight travel delays of other vehicles caused by travel maneuvers of the autonomous vehicle may be tolerated.
In another embodiment of the invention, the traffic condition is reaching an intersection and the driving maneuver to be performed by the autonomous vehicle is driving over the intersection, for example turning at the intersection or traversing the intersection on a lane that the autonomous vehicle has used before reaching the intersection. Intersection refers herein to any intersection of traffic routes. For example, an intersection where one traffic route ends and meets another traffic route ("T-intersection") is also referred to as an intersection. An arrival intersection is a potentially challenging traffic situation for autonomous vehicles because it may be driven in by vehicles from different directions and often relatively complex rules of preference need to be complied with, especially if it is not an intersection controlled, for example, by a traffic light system. The method according to the invention is thus particularly directed to the driving behaviour of an autonomous vehicle when arriving at and driving over an intersection.
In the above-described embodiments of the invention, for example, the following variables are defined and used as a measure of traffic flow:
the individual waiting time for the autonomous vehicle to wait in front of the intersection, whereas it is the first vehicle waiting in front of the intersection on the lane in which it is travelling,
-overall waiting time, representing the total waiting time of vehicles waiting behind the autonomous vehicle before the intersection during the individual waiting time, on the lane on which the autonomous vehicle is travelling, and
-dominant traffic flow, defined as the product of individual waiting time and the number of all vehicles passing through the intersection during the individual waiting time using at least one lane that the autonomous vehicle has to pass through when driving through the intersection or that the autonomous vehicle has to merge.
Individual latency is a measure of the travel delay of an autonomous vehicle. The overall latency is the sum of the latencies of other vehicles waiting at the intersection behind the autonomous vehicle during the individual latencies. Thus, overall latency is a measure of the travel delays of all vehicles that are impeded from continuing to use the lane in which the autonomous vehicle is traveling due to the autonomous vehicle waiting at the intersection during the individual latency. Dominant traffic flow is a measure of traffic flow in all other lanes where vehicles may be blocked by the driving maneuver of an autonomous vehicle.
In the case where the traffic condition is an arrival intersection, to optimize the traffic flow, for example, the self traffic flow ratio is minimized, which is defined as the sum of individual latency and overall latency divided by the dominant traffic flow. According to the definition of the self-traffic flow ratio, the minimization of the self-traffic flow ratio aims at optimizing a balance between the running delay of the autonomous vehicle and the running delays of other vehicles caused by the driving behavior of the autonomous vehicle.
Or, for example, minimizing the self traffic flow ratio under constraint of minimum steering delay. Here, the steering delay is defined as a function of a speed difference caused by running steering for a vehicle braked by executing the running steering, wherein each speed difference caused by running steering is defined as a difference between a reference speed of the braked vehicle and a speed to which the vehicle is braked.
For example, the steering delay is defined as the sum of speed differences caused by driving steering during the evaluation period.
For example, the evaluation period extends from the beginning of the driving maneuver to the latest point in time at which a speed difference caused by the driving maneuver can be detected with the at least one sensor of the autonomous vehicle.
The reference speed of the braked vehicle is, for example, the minimum value of the speed of the braked vehicle that matches the traffic flow immediately before running maneuvers are performed and the maximum allowable speed on the lane used by the braked vehicle. Alternatively, the reference speed of the braked vehicle is, for example, the speed of the braked vehicle immediately before running maneuver is performed.
Drawings
Embodiments of the present invention will be explained in more detail below with reference to the accompanying drawings. Wherein:
figure 1 shows a flow chart of an embodiment of the method according to the invention,
fig. 2 shows a road scene of an intersection.
Detailed Description
Fig. 1 (FIG 1) shows a flow chart 100 of an embodiment of a method of adapting the driving behavior of an autonomous vehicle according to the invention, said method having method steps 101 to 103.
Method steps 101 to 103 will also be described below with reference to fig. 2.
Fig. 2 (FIG 2) schematically shows a road scene 200 at an intersection 201. The first road 202 and the second road 203 intersect at an intersection 201, wherein the first road 202 is a priority road at the intersection 201. The first road 202 has two lanes 204, 205 with mutually different prescribed directions of travel. The second road 203 also has two lanes 206, 207 with mutually different prescribed directions of travel. The respective prescribed travel directions of the lanes 204 to 207 are indicated by arrows in fig. 2.
Also shown are vehicles 208, 209, 210 using the first lane 204 of the first road 202, vehicles 211, 212, 213 using the second lane 205 of the first road 202, vehicles 214 using the first lane 206 of the second road 203, and vehicles 215, 216, 217 using the second lane 207 of the second road 203. Vehicle 217 is an autonomous vehicle.
In a first method step 101, a traffic situation is defined with a driving maneuver to be performed by the autonomous vehicle 217.
The definition of traffic conditions includes, for example, a description of the static structure of the traffic environment. This includes, for example, traffic routes in traffic environments, such as roads and their lanes and their connections, but also crosswalks or traffic lights. In addition, the definition of traffic conditions includes, for example, traffic rules applicable in traffic environments, such as priority rules or speed limits. In addition, the definition of the traffic condition also includes a description of the driving maneuver to be performed by the autonomous vehicle 217, i.e., the driving mission of the autonomous vehicle 217.
In the example shown in fig. 2, the traffic condition is that an autonomous vehicle 217 having a driving mission arrives at the intersection 201. The relevant definition of traffic conditions includes an abstraction of an intersection 201 shown in fig. 2, which has two intersecting roads 202, 203 and their lanes 204 to 207. Furthermore, the definition of traffic conditions includes traffic rules applicable at the intersection 201, in particular indicating that the first road 202 is a priority road, and the travel directions specified for the lanes 204 to 207. Further, the definition of traffic conditions includes a description of the driving task of the autonomous vehicle 217, i.e., the driving maneuver to be performed by the autonomous vehicle 217 at the intersection 201. For example, a driving task is to turn from the second lane 207 of the second road 203 to the first lane 204 of the first road 202 at the intersection 201, to turn from the second lane 207 of the second road 203 to the second lane 205 of the first road 202 at the intersection 201, or to traverse the intersection 201 on the second lane 207 of the second road 203.
In a second method step 102, a model for carrying out driving maneuvers in traffic situations is trained. The model is based on, for example, an artificial neural network.
The model considers and evaluates traffic flows related to traffic conditions. The execution of the driving maneuver is trained in such a way that the traffic flow is optimized according to its evaluation. When evaluating the traffic flow, the travel delays of the autonomous vehicle 217 and the travel delays of the other vehicles 208 to 216 caused by the driving behavior of the autonomous vehicle 217 are taken into account. When optimizing the traffic flow, for example, pareto optimization is performed with the aim of minimizing the travel delay of the autonomous vehicle 217 and the travel delays of the other vehicles 208 to 216 caused by the driving behavior of the autonomous vehicle 217.
Hereinafter, the training of such a model is described using an example of traffic conditions that arrive at the intersection 201 shown in fig. 2, wherein the driving maneuver to be performed by the autonomous vehicle 217 at the intersection 201 is a transition from the lane 207 of the second road 203 to one of the lanes 204, 205 of the first road 202 at the intersection 201 or a crossing of the intersection 201 on the lane 207 of the second road 203.
Here, for example, a stop is defined at which the autonomous vehicle 217 must finally stop on the lane 207 in front of the intersection 201 in order to avoid encroaching on the lane 205 of the first road 202. In fig. 2, the autonomous vehicle 217 is at the stopping point. At this stopping point, the autonomous vehicle 217 generally must stop waiting until it can enter the intersection 201, because, due to the preferential nature of the first road 202, it must first pass vehicles 208 to 213 entering the intersection 201 from one of the lanes 204, 205 of the first road 202 through the intersection 201. At the stopping point, the autonomous vehicle 217 is the first vehicle waiting in front of the intersection 201 on the lane 207 in which it is using. Vehicles 215, 216 traveling behind the autonomous vehicle 217 on the lane 207 must also park and wait until the autonomous vehicle 217 leaves the parking spot.
In particular, training of the model is used to determine a point in time when the autonomous vehicle 217 leaves the parking spot to perform the driving maneuver. The determination of this point in time is such that the traffic flow is optimized by the execution of the driving maneuver as a function of its evaluation.
In order to evaluate and optimize traffic flow, the following variables are defined and used as metrics for traffic flow:
the individual waiting time for the autonomous vehicle 217 waiting at the stop or the individual waiting time for the autonomous vehicle 217 waiting in front of the intersection 201, whereas it is the first vehicle waiting in front of the intersection 201 on the lane 207 in which it is travelling,
overall latency, representing the total latency of vehicles 215, 216 waiting behind autonomous vehicle 217 in front of intersection 201 during individual latencies, i.e., the sum of the latencies of all vehicles 215, 216 waiting behind autonomous vehicle 217 in front of intersection 201 during individual latencies, and
dominant traffic flow, defined as the product of individual waiting time and the number of all vehicles 208 to 214 that pass through the intersection using at least one lane 204 to 206 during the individual waiting time, the autonomous vehicle 217 must pass through or the autonomous vehicle 217 must incorporate at least one of said lanes 204 to 206 while driving through the intersection 201.
In the case of a driving maneuver that is traversing the intersection 201 on the lane 207, the autonomous vehicle 217 must traverse both lanes 204, 205 of the first road 202. In this case, the dominant traffic flow is thus the product of the individual waiting time and the number of all vehicles 208 to 214 that pass the intersection using one of the lanes 204, 205 of the first road 202 during the individual waiting time.
In the event that the travel maneuver is to turn from lane 207 to lane 205, the autonomous vehicle 217 must incorporate lane 205. In this case, the dominant traffic flow is thus the product of the individual waiting time and the number of all vehicles 211 to 213 that pass through the intersection using the lane 205 of the first road 202 during the individual waiting time.
In the event that the driving maneuver is to go from lane 207 to lane 204, the autonomous vehicle must pass through lanes 205 and 206 and merge into lane 204. In this case, the dominant traffic flow is thus the product of the individual waiting time and the number of all vehicles 208 to 214 passing through the intersection using at least one of the lanes 204 to 206 during the individual waiting time.
A self-traffic flow ratio associated with the autonomous vehicle 217 is formed from the individual latency, the overall latency, and the dominant traffic flow, the ratio being defined as the sum of the individual latency and the overall latency divided by the dominant traffic flow. The model performing the driving maneuver is then trained, for example, to minimize the self traffic flow ratio.
Alternatively, the model for performing the driving maneuver is trained, for example, such that the self traffic flow ratio is minimized under the constraint that the maneuver delay is minimized. The steering delay is defined as a function of the difference in speed caused by the running maneuver for the speed of the vehicle 208 to 214 braked as a result of executing the running maneuver, wherein each difference in speed caused by the running maneuver is defined as the difference between the reference speed of the braked vehicle 208 to 214 and the speed to which the vehicle 208 to 214 is braked.
For example, the steering delay is defined as the sum of speed differences caused by driving steering during the evaluation period. For example, the evaluation period extends from the beginning of the driving maneuver to the latest point in time at which a speed difference caused by the driving maneuver can be detected with the at least one sensor of the autonomous vehicle.
The reference speed of the braked vehicles 208 to 214 is, for example, the minimum value of the speed of the braked vehicles 208 to 214 that matches the traffic flow immediately before the running maneuver is performed and the maximum allowable speed on one of the lanes 204 to 206 used by the braked vehicles 208 to 214. Alternatively, the reference speed of the braked vehicles 208 to 214 is the speed of the braked vehicles 208 to 214 immediately before the running maneuver is performed.
In this case, the model thus acts as a classifier of the situation of the autonomous vehicle 217 at the parking spot before the intersection 201. The model determines real-time advice for parking at a parking spot or for driving away from a parking spot as output parameters. The input parameters of the model are descriptions of traffic conditions, including driving maneuvers to be performed and dynamic boundary conditions, such as the current occupancy of the respective lane 204 to 207, the current traffic flow on the lane 204 to 207 or the average vehicle distance on the lane 204 to 207.
In a third method step 103, the model is applied in reality by the autonomous vehicle 217. The autonomous vehicle 217 generally follows a path of travel from a starting location to a target location. For navigation and object detection along a travel route, the autonomous vehicle 217 has a navigation device and a plurality of sensors, such as cameras, lidar sensors, radar sensors and/or ultrasound sensors. The navigation device includes a position determination unit that receives and evaluates satellite signals from navigation satellites of one or more navigation satellite systems, such as GPS, GLONASS, galileo or/and the Beidou navigation satellite system, to determine and track the current position of the autonomous vehicle 217. Further, the navigation device evaluates map data of one or more digital maps stored in a memory unit of the autonomous vehicle 217 and/or provided by a data Cloud service of a data Cloud (english Cloud). Further, the autonomous vehicle 217 may be configured to receive and evaluate the data of the V2X communication. V2X communication is also referred to as vehicle-to-everything communication, vehicle-X communication or vehicle-to-X communication, and is capable of exchanging data between a vehicle and the surroundings of the vehicle, in particular communication between vehicles may also be implemented, which is also referred to as vehicle-to-vehicle communication, V2V communication or vehicle-to-vehicle communication. V2X communication is typically performed by radio signals in the WLAN band, with DSRC (short range communication for dedicated) or mobile wireless networks.
To apply the trained model, the autonomous vehicle 217 continuously acquires dynamic input parameters of the model as it approaches traffic conditions. In the traffic situation described above with reference to fig. 2, the autonomous vehicle 217 detects the current occupancy of the respective lanes 204 to 207, the current speed of the vehicles 208 to 216 on the lanes 204 to 207, and the current vehicle distance on the lanes 204 to 207 when approaching the intersection 201, for example using its sensors. The model provides continuous action advice to the autonomous maneuver planning of the autonomous vehicle 217 based on the dynamic input parameters. In the traffic situation described above with reference to fig. 2, at a parking spot on the lane 207 before the intersection 201, the action advice is, for example, waiting at the parking spot or driving away from the parking spot.
The second method step 102 is first performed during the training phase. Preferably, the model is trained here with data recorded in reality. For example, the model here specifically rewards the execution of driving maneuvers performed in reality by non-autonomous vehicles by reinforcement learning. Thus, the execution of the driving maneuver by the autonomous vehicle 217 may be adapted to the behavior of the non-autonomous vehicle driver. After the training phase, the trained model is transferred to the autonomous vehicle 217.
In a third method step 103, the traffic flow may be detected by the autonomous vehicle 217 during and after the application of the driving maneuver, and the resulting optimization parameters are determined, which are optimized when training the model. In the traffic situation described above with reference to fig. 2, the autonomous vehicle 217 detects, for example, its individual waiting time, overall waiting time, dominant traffic flow, and speed differences of the other vehicles 208 to 213 caused by the driving maneuver, and thereby determines the self traffic flow ratio and the maneuver delay as optimization parameters.
The optimization parameters determined during and after the application of the driving maneuver may then be used to further train the model, which may result in an improvement of the model. Further training of the model may be performed, for example, by an in-vehicle unit of the autonomous vehicle 217. Alternatively or additionally, the determined optimization parameters are sent, for example, to a data cloud that manages one instance of the model and continuously trains and improves the model based on data sent to it by other vehicles. Examples of models trained in the data cloud may be used, for example, to implement it in the autonomous vehicle 217, or to replace or update examples of models already implemented in the autonomous vehicle 217.
List of reference numerals
100. Flow chart
101 to 103 method steps
200. Road scene
201. Intersection of crossing
202. 203 road
204 to 207 lanes
208 to 216 vehicle
217. Autonomous vehicle
Claims (15)
1. A method of adapting driving behavior of an autonomous vehicle (217), wherein
Defining a traffic condition comprising a driving maneuver to be performed by the autonomous vehicle (217),
training a model for performing driving maneuvers in the traffic situation,
wherein the model considers and evaluates traffic flows related to the traffic conditions and trains the execution of driving maneuvers in a manner that optimizes traffic flows according to its evaluation,
-and under said traffic conditions, performing a driving maneuver by the autonomous vehicle (217) depending on the traffic flow according to said model.
2. The method of claim 1, wherein the model is based on an artificial neural network.
3. The method according to claim 1 or 2, wherein the model is trained with data recorded in reality.
4. A method according to claim 3, wherein the model specifically rewards execution of a driving maneuver that has been executed in reality by a non-autonomous vehicle by reinforcement learning.
5. The method according to any of the preceding claims, wherein the traffic flow in the traffic condition is detected with at least one sensor of an autonomous vehicle (217).
6. The method according to any of the preceding claims, wherein the travel delay of the autonomous vehicle (217) and the travel delay of the other vehicles (208 to 216) caused by the driving behaviour of the autonomous vehicle (217) are taken into account when evaluating the traffic flow.
7. The method of claim 6, wherein pareto optimization is performed while optimizing traffic flow with the objective of minimizing travel delays of the autonomous vehicle (217) and of other vehicles (208 to 216) caused by driving behavior of the autonomous vehicle (217).
8. The method according to any of the preceding claims, wherein the traffic condition is reaching an intersection (201) and the driving maneuver is driving over the intersection (201).
9. The method of claim 8, wherein the following variables are defined and used as metrics for traffic flow:
an individual waiting time for the autonomous vehicle (219) to wait in front of the intersection (201), whereas it is the first vehicle waiting in front of the intersection (201) on the lane (207) in which it is travelling,
-an overall waiting time, which represents the total waiting time of vehicles (215, 216) waiting behind the autonomous vehicle (217) in front of the intersection (201) during the individual waiting time on the lane (207) on which the autonomous vehicle (217) is travelling, and
-a dominant traffic flow defined as the product of the individual waiting time and the number of all vehicles (208 to 214) passing through the intersection (201) using at least one lane (204 to 206) that the autonomous vehicle (217) has to traverse when driving through the intersection (201) or that the autonomous vehicle (217) has to merge into during the individual waiting time.
10. The method of claim 9, wherein to optimize traffic flow, the self traffic flow ratio is minimized, defined as the sum of individual latency and overall latency divided by the dominant traffic flow.
11. The method of claim 10, wherein the self traffic flow ratio is minimized under the constraint of minimum steering delay, wherein steering delay is defined as a function of a speed difference caused by a driving maneuver for a vehicle (208 to 214) braked by performing the driving maneuver, wherein each driving maneuver caused speed difference is defined as a difference between a reference speed of the braked vehicle (208 to 214) and a speed at which the vehicle (208 to 214) is braked.
12. The method of claim 11, wherein the maneuver delays are defined as a sum of speed differences caused by a driving maneuver during the evaluation period.
13. The method of claim 12, wherein the evaluation period extends from a beginning of a driving maneuver to a latest point in time at which a speed difference caused by the driving maneuver is detectable with at least one sensor of the autonomous vehicle (217).
14. The method according to any one of claims 11 to 13, wherein the reference speed of the braked vehicle (208 to 214) is the minimum of the speed of the braked vehicle (208 to 214) that matches the traffic flow immediately before the running maneuver is performed and the maximum allowable speed on the lane (204 to 206) used by the braked vehicle (208 to 214).
15. The method according to any one of claims 11 to 13, wherein the reference speed of the braked vehicle (208 to 214) is the speed of the braked vehicle (208 to 214) immediately before the running maneuver is performed.
Applications Claiming Priority (2)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
DE102022106338.0 | 2022-03-18 | ||
DE102022106338.0A DE102022106338A1 (en) | 2022-03-18 | 2022-03-18 | Adapting driving behavior of an autonomous vehicle |
Publications (1)
Publication Number | Publication Date |
---|---|
CN116767272A true CN116767272A (en) | 2023-09-19 |
Family
ID=87849613
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN202310196851.7A Pending CN116767272A (en) | 2022-03-18 | 2023-03-03 | Adaptation of driving behavior of autonomous vehicles |
Country Status (2)
Country | Link |
---|---|
CN (1) | CN116767272A (en) |
DE (1) | DE102022106338A1 (en) |
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114179801A (en) * | 2020-08-24 | 2022-03-15 | 均联智行有限公司 | Determining an optimal distance between two vehicles |
Family Cites Families (3)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
EP3598414A1 (en) | 2018-07-20 | 2020-01-22 | Volvo Car Corporation | System and method for avoiding a collision course |
US11861458B2 (en) | 2018-08-21 | 2024-01-02 | Lyft, Inc. | Systems and methods for detecting and recording anomalous vehicle events |
DE102020201931A1 (en) | 2020-02-17 | 2021-08-19 | Psa Automobiles Sa | Method for training at least one algorithm for a control unit of a motor vehicle, method for optimizing a traffic flow in a region, computer program product and motor vehicle |
-
2022
- 2022-03-18 DE DE102022106338.0A patent/DE102022106338A1/en active Pending
-
2023
- 2023-03-03 CN CN202310196851.7A patent/CN116767272A/en active Pending
Cited By (1)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN114179801A (en) * | 2020-08-24 | 2022-03-15 | 均联智行有限公司 | Determining an optimal distance between two vehicles |
Also Published As
Publication number | Publication date |
---|---|
DE102022106338A1 (en) | 2023-09-21 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
RU2671446C1 (en) | Autonomous driving system and the autonomous driving vehicle | |
CN111696373B (en) | Motorcade cooperative sensing method, motorcade cooperative control method and motorcade cooperative control system | |
CN111448597B (en) | Driving control method and driving control device | |
JP7347523B2 (en) | Vehicle control device and vehicle control method | |
IL293713A (en) | Navigation based on vehicle activity | |
JP7331939B2 (en) | In-vehicle device and driving support method | |
KR20190030199A (en) | Supervision of vehicles | |
JP7409810B2 (en) | Cooperative operation coordination method | |
GB2524384A (en) | Autonomous control in a dense vehicle environment | |
US20200272159A1 (en) | Method and vehicle control system for intelligent vehicle control about a roundabout | |
US11615702B2 (en) | Determining vehicle path | |
US20170349181A1 (en) | Lane management system for an automated vehicle | |
JP5565303B2 (en) | Driving support device and driving support method | |
US10144428B2 (en) | Traffic light operation | |
US11685398B2 (en) | Lane based routing system for autonomous driving vehicles | |
US11358606B2 (en) | Method for operating a motor vehicle for improving working conditions of evaluation units in the motor vehicle, control system for performing a method of this kind, and motor vehicle having a control system of this kind | |
CN115701295A (en) | Method and system for vehicle path planning | |
JP7362899B2 (en) | Vehicle driving support method and driving support device | |
US20210188273A1 (en) | Enhanced vehicle operation | |
JP7452650B2 (en) | Parking/stopping point management device, parking/stopping point management method, vehicle device | |
CN115379977A (en) | Method for planning a theoretical movement trajectory | |
JP7400754B2 (en) | Remote support system and remote support method | |
US20230174106A1 (en) | Path checking device and path checking method | |
WO2021115874A1 (en) | Method for controlling an ego vehicle in an environment of a plurality of gateways | |
CN116767272A (en) | Adaptation of driving behavior of autonomous vehicles |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination |