CN116448134A - Vehicle path planning method and device based on risk field and uncertain analysis - Google Patents
Vehicle path planning method and device based on risk field and uncertain analysis Download PDFInfo
- Publication number
- CN116448134A CN116448134A CN202310323365.7A CN202310323365A CN116448134A CN 116448134 A CN116448134 A CN 116448134A CN 202310323365 A CN202310323365 A CN 202310323365A CN 116448134 A CN116448134 A CN 116448134A
- Authority
- CN
- China
- Prior art keywords
- target
- risk field
- label
- prediction
- speed
- 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.)
- Granted
Links
- 238000000034 method Methods 0.000 title claims abstract description 65
- 238000004458 analytical method Methods 0.000 title abstract description 9
- 230000008447 perception Effects 0.000 claims abstract description 79
- 238000004364 calculation method Methods 0.000 claims abstract description 26
- 238000011478 gradient descent method Methods 0.000 claims abstract description 19
- 238000012805 post-processing Methods 0.000 claims abstract description 15
- 230000015654 memory Effects 0.000 claims description 43
- 238000013527 convolutional neural network Methods 0.000 claims description 30
- 238000013076 uncertainty analysis Methods 0.000 claims description 21
- 230000006870 function Effects 0.000 claims description 17
- 238000011156 evaluation Methods 0.000 claims description 13
- 238000012549 training Methods 0.000 claims description 12
- 238000011176 pooling Methods 0.000 claims description 11
- 238000010276 construction Methods 0.000 claims description 9
- 238000004590 computer program Methods 0.000 claims description 7
- 230000008859 change Effects 0.000 claims description 4
- 238000000605 extraction Methods 0.000 claims description 3
- 238000004422 calculation algorithm Methods 0.000 abstract description 29
- 238000010586 diagram Methods 0.000 description 10
- 238000004891 communication Methods 0.000 description 8
- 230000008569 process Effects 0.000 description 8
- 238000001514 detection method Methods 0.000 description 6
- 238000004088 simulation Methods 0.000 description 6
- 230000008901 benefit Effects 0.000 description 5
- 239000003795 chemical substances by application Substances 0.000 description 5
- 238000005290 field theory Methods 0.000 description 5
- 238000012545 processing Methods 0.000 description 5
- 238000013528 artificial neural network Methods 0.000 description 4
- 230000001133 acceleration Effects 0.000 description 3
- 238000013135 deep learning Methods 0.000 description 3
- 239000011159 matrix material Substances 0.000 description 3
- 238000005381 potential energy Methods 0.000 description 3
- 238000003491 array Methods 0.000 description 2
- 238000005516 engineering process Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 239000000463 material Substances 0.000 description 2
- 230000003287 optical effect Effects 0.000 description 2
- 238000013138 pruning Methods 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 238000005070 sampling Methods 0.000 description 2
- 238000010845 search algorithm Methods 0.000 description 2
- 238000013459 approach Methods 0.000 description 1
- 230000007547 defect Effects 0.000 description 1
- 238000000802 evaporation-induced self-assembly Methods 0.000 description 1
- 238000002474 experimental method Methods 0.000 description 1
- 238000009472 formulation Methods 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 239000000203 mixture Substances 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000010606 normalization Methods 0.000 description 1
- 239000013307 optical fiber Substances 0.000 description 1
- 230000002093 peripheral effect Effects 0.000 description 1
- 230000002265 prevention Effects 0.000 description 1
- 230000002035 prolonged effect Effects 0.000 description 1
- 230000002787 reinforcement Effects 0.000 description 1
- 230000004044 response Effects 0.000 description 1
- 230000002441 reversible effect Effects 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 230000006403 short-term memory Effects 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000005728 strengthening Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/26—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 specially adapted for navigation in a road network
- G01C21/34—Route searching; Route guidance
- G01C21/3407—Route searching; Route guidance specially adapted for specific applications
-
- 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
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/10—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00 by using measurements of speed or acceleration
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01C—MEASURING DISTANCES, LEVELS OR BEARINGS; SURVEYING; NAVIGATION; GYROSCOPIC INSTRUMENTS; PHOTOGRAMMETRY OR VIDEOGRAMMETRY
- G01C21/00—Navigation; Navigational instruments not provided for in groups G01C1/00 - G01C19/00
- G01C21/20—Instruments for performing navigational calculations
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/044—Recurrent networks, e.g. Hopfield networks
- G06N3/0442—Recurrent networks, e.g. Hopfield networks characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
- G06N3/0455—Auto-encoder networks; Encoder-decoder networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/0464—Convolutional networks [CNN, ConvNet]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/047—Probabilistic or stochastic networks
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
- G06N3/084—Backpropagation, e.g. using gradient descent
-
- Y—GENERAL TAGGING OF NEW TECHNOLOGICAL DEVELOPMENTS; GENERAL TAGGING OF CROSS-SECTIONAL TECHNOLOGIES SPANNING OVER SEVERAL SECTIONS OF THE IPC; TECHNICAL SUBJECTS COVERED BY FORMER USPC CROSS-REFERENCE ART COLLECTIONS [XRACs] AND DIGESTS
- Y02—TECHNOLOGIES OR APPLICATIONS FOR MITIGATION OR ADAPTATION AGAINST CLIMATE CHANGE
- Y02T—CLIMATE CHANGE MITIGATION TECHNOLOGIES RELATED TO TRANSPORTATION
- Y02T10/00—Road transport of goods or passengers
- Y02T10/10—Internal combustion engine [ICE] based vehicles
- Y02T10/40—Engine management systems
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Radar, Positioning & Navigation (AREA)
- Remote Sensing (AREA)
- General Physics & Mathematics (AREA)
- Molecular Biology (AREA)
- Data Mining & Analysis (AREA)
- Software Systems (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Mathematical Physics (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Computing Systems (AREA)
- Automation & Control Theory (AREA)
- Human Computer Interaction (AREA)
- Transportation (AREA)
- Mechanical Engineering (AREA)
- Probability & Statistics with Applications (AREA)
- Traffic Control Systems (AREA)
Abstract
The application relates to a vehicle path planning method and device based on a risk field and uncertain analysis, wherein the method comprises the following steps: performing label classification and position speed prediction on the perceived target to obtain label classification and position speed prediction results of the perceived target; based on label classification and position speed prediction results of a perception target, respectively establishing a label probability risk field and a prediction dynamic risk field for measuring uncertainty in label classification and position speed prediction; and acquiring a planning trace cluster by a gradient descent method in each preset time step based on the tag probability risk field and the predicted dynamic risk field, and performing post-processing on the planning trace cluster to select an optimal path. Thereby solving the problems that the existing path planning decision algorithm considering uncertainty has too high calculation power requirement and poor robustness, and is difficult to apply to complex scenes, etc.
Description
Technical Field
The present disclosure relates to the field of path planning technologies, and in particular, to a vehicle path planning method and apparatus based on risk fields and uncertain analysis.
Background
The existing unmanned path planning algorithm can be divided into a graph searching method, a random sampling method, a geometric curve method and an optimizing method, wherein in the graph searching method, an algorithm A searches outwards by taking a starting point as a center, takes a passing path as a cost value and takes a linear distance between the starting point and a terminal point as a heuristic value, so that the searching is performed along the shortest distance direction, the algorithm can plan an optimal path, but the incremental calculation method of the algorithm causes high calculation complexity; the rapid random search tree algorithm generates an expansion tree in a random sampling mode, searches a planning path in the expansion tree, and has high speed, wherein the planning path is a local optimal track; the geometric curve method is formed by fitting approximation control points, has good smoothness, can adapt to vehicle dynamics constraint conditions, but is difficult to avoid obstacles in real time; the optimal control method finds a feasible optimal control amount through an optimal control theory, and has the defect that the characteristics of an optimal track are difficult to describe accurately. However, most of these existing path planning algorithms do not take uncertainty factors into consideration, and it is difficult to give a correct decision in case of ambiguity of the perception side information.
In the existing planning algorithm considering uncertainty, uncertainty of environment perception information is reflected on map modeling by a path planning algorithm based on azimuth relation, a feature map with probability distribution is constructed to carry out convex polygon subdivision, the searching performance of Dijkstra algorithm is improved by utilizing direction information of subdivision blocks, and meanwhile safety of planning paths under the condition of environment perception uncertainty is guaranteed, so that efficiency of the Dijkstra algorithm is improved, and an optimal safety path can be planned under the influence of uncertainty; gradually planning a track diagram on the probability distribution space of the state based on a random tree search algorithm, searching candidate paths when new samples are added, pruning by utilizing the propagation characteristics of uncertainty, and thus strengthening the convergence of the algorithm by utilizing a new pruning technology; the ant colony algorithm-based planning method utilizes detection information in the advancing process to complete local prediction and collision prevention planning, can adapt to dynamically-changing environments, and the improved ant colony algorithm converges faster, so that uncertainty caused by obstacle movement is well solved.
However, the path planning algorithm based on the azimuth relation cannot process an excessively complex environment, the operation performance is insufficient when the obstacle is increased, the calculation force requirement of the planning algorithm based on the random tree search algorithm is excessively high, the calculation time fluctuation is large in practical application, the perception end uncertainty is not considered in the planning based on the improved ant colony algorithm, and meanwhile, the robustness of the path planning method is poor and needs to be solved urgently.
Disclosure of Invention
The application provides a vehicle path planning method and device based on a risk field and uncertainty analysis, which are used for solving the problems that the existing path planning decision algorithm considering uncertainty is too high in calculation power requirement, poor in robustness, difficult to apply to complex scenes and the like.
An embodiment of a first aspect of the present application provides a vehicle path planning method based on risk fields and uncertainty analysis, including the following steps: performing label classification and position speed prediction on the perceived target to obtain label classification and position speed prediction results of the perceived target; based on the label classification and position speed prediction results of the perception target, respectively establishing a label probability risk field and a prediction dynamic risk field for measuring uncertainty in the label classification and position speed prediction; and acquiring a planning trace cluster by a gradient descent method in each preset time step based on the tag probability risk field and the predicted dynamic risk field, and performing post-processing on the planning trace cluster to select an optimal path.
Optionally, in an embodiment of the present application, the performing tag classification and location speed prediction on the perceived target to obtain a tag classification and location speed prediction result of the perceived target includes: constructing a multi-layer three-channel convolutional neural network, and training the multi-layer three-channel convolutional neural network; based on preset conditions, removing the last layer of the trained multi-layer three-channel convolutional neural network, and outputting the probability that the perception target belongs to each label; extracting time information of the perception target by using a long-period memory network encoder, and collecting long-period memory network states of all surrounding intelligent agents in a tensor in a social pooling layer; extracting the space information of the perception target from the convolutional neural network; and generating probability distribution of the perception target in the transverse working condition and the longitudinal working condition by using a preset number of long-short-term memory network decoders so as to obtain probability distribution prediction of the track and the speed of the perception target.
Optionally, in an embodiment of the present application, the establishing a tag probability risk field and a predicted dynamic risk field for measuring uncertainty in the tag classification and the position velocity prediction based on the tag classification and the position velocity prediction result of the perceived target includes: when the tag information meets the preset condition, measuring the risk of the obstacle based on the tag probability risk field; when the target speed information meets the preset condition, measuring the target risk size based on a predicted dynamic risk field; and establishing a probability dynamic risk field based on the tag probability risk field and a prediction dynamic risk field to measure the risk when the tag classification and the position speed prediction are uncertain.
Optionally, in an embodiment of the present application, the acquiring a planned trace cluster by a gradient descent method in each preset time step based on the tag probability risk field and the predicted dynamic risk field, and performing post-processing on the planned trace cluster to select an optimal path includes: calculating repulsive force of a vehicle position at the current moment, defining attractive force which always points to a target point position from the vehicle position, obtaining resultant force of the repulsive force and the attractive force at the current moment through linear superposition, obtaining speed at the current moment, and obtaining planning speed at the next moment based on the speed at the current moment and the resultant force; iteratively calculating repulsive force, attractive force and planning speed at the vehicle position based on preset iteration times, and obtaining a plurality of vehicle position coordinates to form a planning track; continuously changing the preset super-parameter sizes of the repulsive force and the attractive force to obtain candidate track clusters; and inputting the candidate track clusters into a preset evaluation function module, and calculating the score of each track, wherein the track corresponding to the highest score item is the optimal path.
Optionally, in an embodiment of the present application, a calculation formula of the probabilistic dynamic risk field is as follows:
Wherein V is i For the risk fields of the perceived objects appearing in all the own vehicle vision, i is the number of the perceived object, P ij For each probability of each label of each perceived target, j is the label number, M (r i ,θ i ) The risk field quality of the target and the target label thereof is perceived; r is (r) i To sense the distance between the target and the own vehicle, v i For each perceived target and label corresponding speed, θ i For each included angle formed by the perception target, the target label and the running speed direction of the own vehicle, k i Is a coefficient to be determined.
An embodiment of a second aspect of the present application provides a vehicle path planning apparatus based on risk field and uncertainty analysis, including: the prediction module is used for carrying out label classification and position speed prediction on the perception target to obtain label classification and position speed prediction results of the perception target; the construction module is used for respectively establishing a tag probability risk field and a prediction dynamic risk field for measuring uncertainty in tag classification and position speed prediction based on the tag classification and position speed prediction result of the perception target; and the selecting module is used for acquiring a planning trace cluster through a gradient descent method in each preset time step based on the tag probability risk field and the prediction dynamic risk field, and carrying out post-processing on the planning trace cluster so as to select an optimal path.
Optionally, in one embodiment of the present application, the prediction module includes: the training unit is used for constructing a multi-layer three-channel convolutional neural network and training the multi-layer three-channel convolutional neural network; the removing unit is used for removing the last layer of the trained multi-layer three-channel convolutional neural network based on preset conditions and outputting the probability that the perception target belongs to each label; the collecting unit is used for extracting the time information of the perception target by utilizing the long-period memory network encoder and collecting the long-period memory network states of all surrounding intelligent agents into a tensor in a social pooling layer; the extraction unit is used for extracting the space information of the perception target through a convolutional neural network; the generation unit is used for generating probability distribution of the perception target in the transverse working condition and the longitudinal working condition by utilizing the preset number of long-short-term memory network decoders so as to obtain probability distribution prediction of the track and the position speed of the perception target.
Optionally, in one embodiment of the present application, the building block includes: the first measuring unit is used for measuring the risk of the obstacle based on the tag probability risk field when the tag information meets the preset condition; the second measuring unit is used for measuring the target risk size based on the predicted dynamic risk field when the target speed information meets the preset condition; and the third measuring unit is used for establishing a probability dynamic risk field based on the tag probability risk field and the prediction dynamic risk field so as to measure the risk when the tag classification and the position speed prediction are uncertain.
Optionally, in one embodiment of the present application, the selecting module includes: the first calculation unit is used for calculating repulsive force at the vehicle position at the current moment, defining attractive force which is always pointed to the target point position by the vehicle position, obtaining resultant force of the repulsive force and the attractive force at the current moment through linear superposition, obtaining the speed at the current moment, and obtaining planning speed at the next moment based on the speed at the current moment and the resultant force; the second calculation unit is used for iteratively calculating repulsive force, attractive force and planning speed at the vehicle position based on preset iteration times, and obtaining a plurality of vehicle position coordinates to form a planning track; a changing unit, configured to continuously change the repulsive force and the preset super-parameter of the attractive force to obtain a candidate track cluster; and the third calculation unit is used for inputting the candidate track clusters into a preset evaluation function module, calculating the score of each track, and taking the track corresponding to the highest score item as the optimal path.
Optionally, in an embodiment of the present application, a calculation formula of the probabilistic dynamic risk field is as follows:
wherein V is i For the risk fields of the perceived objects appearing in all the own vehicle vision, i is the number of the perceived object, P ij For each probability of each label of each perceived target, j is the label number, M (r i ,θ i ) The risk field quality of the target and the target label thereof is perceived; r is (r) i To sense the distance between the target and the own vehicle, v i For each perceived target and label corresponding speed, θ i For each included angle formed by the perception target, the target label and the running speed direction of the own vehicle, k i Is a coefficient to be determined.
An embodiment of a third aspect of the present application provides an electronic device, including: the system comprises a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor executes the program to realize the vehicle path planning method based on the risk field and the uncertainty analysis.
A fourth aspect of the present application provides a computer readable storage medium storing a computer program which when executed by a processor implements a method of vehicle path planning based on risk-profile and uncertainty analysis as above.
Thus, embodiments of the present application have the following benefits:
according to the embodiment of the application, the label classification and the position speed prediction result of the perception target can be obtained by carrying out label classification and position speed prediction on the perception target; based on label classification and position speed prediction results of a perception target, respectively establishing a label probability risk field and a prediction dynamic risk field for measuring uncertainty in label classification and position speed prediction; and acquiring a planning trace cluster by a gradient descent method in each preset time step based on the tag probability risk field and the predicted dynamic risk field, and performing post-processing on the planning trace cluster to select an optimal path. According to the method, uncertainty factors are measured through a travelling risk field theory, and then an optimal path is selected according to the established probability dynamic risk field by using a gradient descent method, so that uncertain information of a sensing end in a special environment is fully utilized, robustness of decision control is improved, and accurate decision can be still carried out under the condition that the information of the sensing end has uncertain characteristics. Therefore, the problems that the existing path planning decision algorithm considering uncertainty is too high in calculation power requirement and poor in robustness, and is difficult to apply to complex scenes and the like are solved.
Additional aspects and advantages of the application will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the application.
Drawings
The foregoing and/or additional aspects and advantages of the present application will become apparent and readily appreciated from the following description of the embodiments, taken in conjunction with the accompanying drawings, in which:
FIG. 1 is a flow chart of a vehicle path planning method based on risk fields and uncertainty analysis provided in accordance with an embodiment of the present application;
FIG. 2 is a schematic diagram of a field formulation and field strength of a tag probability risk field according to an embodiment of the present disclosure;
FIG. 3 is a schematic diagram of a field construction formula and field strength for predicting a dynamic risk field according to an embodiment of the present disclosure;
FIG. 4 is a schematic diagram of a gradient planning algorithm according to an embodiment of the present application;
FIG. 5 is a schematic diagram of a trace cluster generation simulation provided in one embodiment of the present application;
FIG. 6 is a schematic diagram of an algorithm planning track under a lane change condition according to an embodiment of the present disclosure;
FIG. 7 is an example diagram of a vehicle path planning apparatus based on risk fields and uncertainty analysis in accordance with an embodiment of the present application;
fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present application.
Wherein, 10-a vehicle path planning device based on risk fields and uncertain analysis, 100-a prediction module, 200-a construction module, 300-an acquisition module, 801-a memory, 802-a processor and 803-a communication interface.
Detailed Description
Embodiments of the present application are described in detail below, examples of which are illustrated in the accompanying drawings, wherein the same or similar reference numerals refer to the same or similar elements or elements having the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and intended for the purpose of explaining the present application and are not to be construed as limiting the present application.
The following describes a vehicle path planning method and device based on risk fields and uncertainty analysis according to an embodiment of the present application with reference to the accompanying drawings. In order to solve the problems mentioned in the background art, the application provides a vehicle path planning method based on a risk field and uncertain analysis, wherein in the method, label classification and position speed prediction results of a perceived target are obtained by carrying out label classification and position speed prediction on the perceived target; based on label classification and position speed prediction results of a perception target, respectively establishing a label probability risk field and a prediction dynamic risk field for measuring uncertainty in label classification and position speed prediction; and acquiring a planning trace cluster by a gradient descent method in each preset time step based on the tag probability risk field and the predicted dynamic risk field, and performing post-processing on the planning trace cluster to select an optimal path. According to the method, uncertainty factors are measured through a travelling risk field theory, and then an optimal path is selected according to the established probability dynamic risk field by using a gradient descent method, so that uncertain information of a sensing end in a special environment is fully utilized, robustness of decision control is improved, and accurate decision can be still carried out under the condition that the information of the sensing end has uncertain characteristics. Therefore, the problems that the existing path planning decision algorithm considering uncertainty is too high in calculation power requirement and poor in robustness, and is difficult to apply to complex scenes and the like are solved.
Specifically, fig. 1 is a flowchart of a vehicle path planning method based on risk field and uncertainty analysis according to an embodiment of the present application.
As shown in fig. 1, the vehicle path planning method based on the risk field and the uncertainty analysis comprises the following steps:
in step S101, label classification and position velocity prediction are performed on the perceived target, and a label classification and position velocity prediction result of the perceived target is obtained.
The embodiment of the application can firstly classify the detection target, namely the perception target, and predict the position and the speed of the detection target, so that the perception information is processed, and a data basis is provided for the construction of a subsequent risk field.
Optionally, in an embodiment of the present application, performing tag classification and location speed prediction on the perceived target to obtain a tag classification and location speed prediction result of the perceived target includes: constructing a multi-layer three-channel convolutional neural network, and training the multi-layer three-channel convolutional neural network; based on preset conditions, removing the last layer of the trained multi-layer three-channel convolutional neural network, and outputting the probability that the perception target belongs to each label; extracting time information of a perception target by using a long-period memory network encoder, and collecting long-period memory network states of all surrounding intelligent agents in a tensor in a social pooling layer; extracting the space information of a perception target from a convolutional neural network; and generating probability distribution of the perception target in the transverse working condition and the longitudinal working condition by using a preset number of long-short-term memory network decoders so as to obtain probability distribution prediction of the track and the speed of the perception target.
According to the embodiment of the application, three-channel raster image sensing data on the github can be selected as samples and are divided into two types of labels without labels and labels, a convolutional neural network CNN is constructed by adopting a target detection method at present, image features are extracted layer by layer in a convolutional-pooling mode, response is increased by using a deep learning method, and more complex and advanced features are extracted.
Specifically, the specific steps for obtaining the target tag classification in the embodiment of the present application are as follows:
(1) A multi-layer three-channel convolutional neural network is constructed, each layer of network is provided with a convolutional layer and a pooling layer, wherein the convolutional layer adopts N convolutional kernels with step length of 1 and 3*3, surrounding data are filled with 0, and the pooling mode adopts maximum pooling so as to introduce nonlinear parts and increase robustness. The penultimate layer adopts a perception machine of four nodes of 'person-vehicle-bicycle-empty', the probability of corresponding labels is to be obtained after normalization, and the last layer adopts a mode of maximum probability to obtain the perception labels;
(2) And initializing parameters of a convolution layer during each training, expanding a convolution kernel matrix and an image data matrix of each channel by adopting an in2col function in NumPy, performing matrix operation by adopting a GPU parallel computing mode, and setting a loss function proportional to the label classification error number. During each training, inputting three-channel raster data with labels in small batches into a neural network, and carrying out back propagation learning parameters according to a loss function to finally train out a network;
(3) And removing the maximum probability value of the last layer of the trained CNN neural network, directly outputting the label probability of the target belonging to each label, and giving the label probability to a decision layer for processing.
After the target label classification is acquired, the embodiment of the application can also predict the position and the speed of the detection target.
The existing track prediction can be divided into four types of methods, namely a physical model-based method, a classical machine learning method, reinforcement learning method and deep learning method, and the embodiment of the application can predict the target position speed in a convolution and recurrent neural network combined mode based on the track prediction method of the deep learning method.
The recurrent neural network can extract time sequence information, the convolutional neural network can extract spatial characteristics considering interaction factors of traffic participants, and the two can make full use of the space-time information of the vehicle in the previous time step to conduct track prediction, so that probability distribution of future motion tracks of the vehicle is obtained, multi-mode track prediction is finally output, the uncertainty corresponding to the embodiment of the application is met, and therefore the embodiment of the application can realize position and speed prediction of a detection target through the following steps:
(1) Extracting time information in the motion information of the other vehicle through an LSTM (Long Short Term Memory network long-short-time memory network) encoder, and collecting LSTM states of all surrounding agents in a tensor in a social pooling layer;
(2) Extracting space related information of the other vehicle by using a convolutional neural network;
(3) Probability distribution conditions of the vehicle in transverse working conditions (left turn, right turn and straight running) and longitudinal working conditions (braking, accelerating and uniform speed) are generated through 6 LSTM decoders, so that probability distribution prediction of the track and speed of the vehicle is obtained.
Therefore, the embodiment of the application classifies the target labels by acquiring and predicts the target position and speed, so that reliable sensing end data support is provided for the establishment of the probability dynamic risk field.
In step S102, a tag probability risk field and a predicted dynamic risk field, which measure uncertainty in tag classification and position velocity prediction, are established based on the tag classification and position velocity prediction results of the perceived target, respectively.
After the label classification and the position speed prediction are performed on the perceived target to obtain the label classification and the position speed prediction result of the perceived target, further, according to the label classification and the position speed prediction result, the embodiment of the application can respectively establish two novel risk fields, namely a label probability risk field and a prediction dynamic risk field, so as to measure the uncertainty in the label classification and the position speed prediction of the perceived end.
Optionally, in one embodiment of the present application, establishing a tag probability risk field and a location velocity prediction risk field measuring uncertainty in tag classification and location velocity prediction, respectively, based on the tag classification and location velocity prediction results of the perceived target, includes: when the tag information meets the preset condition, measuring the risk of the obstacle based on the tag probability risk field; when the target speed information meets the preset condition, measuring the target risk size based on the predicted dynamic risk field; and establishing a probability dynamic risk field based on the tag probability risk field and the prediction dynamic risk field to measure the risk when the tag classification and the position speed prediction are uncertain.
It should be noted that, in the embodiment of the present application, when tag information has uncertainty, the size of the obstacle risk may be measured by the tag probability risk field, and after the target tag probability distribution is obtained, the embodiment of the present application may respectively establish the tag probability risk field for each perceived target.
In the embodiment of the application, a risk field is established for a moving object, and the following three factors are mainly considered to measure the risk brought by the object in driving:
(1) During driving, the severity of the collision of the vehicle with the dynamic obstacle is related to the attribute of the dynamic obstacle, and the collision of the larger obstacle with the smaller obstacle is more serious;
(2) The closer the distance between the running vehicle and the obstacle is, the greater the probability of collision is, and the probability of collision and the loss caused by collision are not increased linearly along with the approach of the distance;
(3) The potential risk that a moving object poses to a running vehicle is related not only to the distance between the running vehicle and the moving object, but also to the speed of the moving object and the position vector of the moving object relative to the running vehicle.
Based on the above factors, embodiments of the present application may establish a tag probability risk field as shown in fig. 2 by:
wherein V is i A risk field representing perceived objects appearing in all the vehicle's views, where i represents the number of the object; p (P) ij Probability (i represents a perceived target number, j represents a label number) for each label of each perceived target; m is M ij Representing the risk field quality of a specific certain perception target and a perception label thereof; r is (r) i Representing the distance between the perception target and the own vehicle, v i Representing the corresponding speed of each perception target and the perception label, theta i Represents the included angle k between each perception target and the perception label and the driving speed direction of the own vehicle i Is a coefficient to be determined.
In addition, when the target speed information has uncertainty, the target risk size can be measured by establishing a predictive dynamic risk field.
After obtaining the speed and position information of the next moment of the target, the embodiment of the application can establish a predicted dynamic risk field of each perceived target through the formula (2), wherein the consideration of the risk brought by the dynamic obstacle in the driving process is consistent with the formula (1):
wherein k is i For undetermined coefficient, r i Representing the distance between the perception target and the own vehicle, theta i Representing the included angle between each perception target and the perception label and the driving speed direction of the own vehicle, v i Representing the corresponding speed of each perceived target and perceived label, M (r i ,θ i ) The mass blocks which are allocated and prolonged are represented, the predicted track of the other vehicle in the next time domain is mainly considered in construction, each predicted point on the track is weighted by mass according to importance, the construction mode of the predicted dynamic risk field can be used for describing the prediction uncertainty of the track of the other vehicle in the next time domain,as shown in fig. 3.
In the actual implementation process, uncertainty is often brought to target label classification and target position and speed prediction, and at this time, the embodiment of the application can establish a probability dynamic risk field by combining a label probability risk field and a prediction dynamic risk field, so that comprehensive measurement of risk size is realized when label classification and position and speed prediction are uncertain.
Optionally, in one embodiment of the present application, the calculation formula of the probabilistic dynamic risk field is as follows:
wherein V is i The risk field of the perception target appearing in the vision of all the vehicles, i is the number of the perception target and P ij For each probability of each label of each perceived target, j is the label number, M (r i ,θ i ) The risk field quality of the target and the target label thereof is perceived; r is (r) i To sense the distance between the target and the own vehicle, v i For each perceived target and label corresponding speed, θ i For each included angle formed by the perception target, the target label and the running speed direction of the own vehicle, k i Is a coefficient to be determined.
It should be noted that, in the embodiment of the present application, a probabilistic dynamic risk field may be constructed by combining the tag probabilistic risk field of the formula (1) and the predictive dynamic risk field of the formula (2), and the probabilistic dynamic risk field is defined as follows:
wherein V is i A risk field representing perceived objects appearing in all the vehicle's views, where i represents the number of the object; p (P) ij Probability (i represents a perceived target number, j represents a label number) for each label of each perceived target; m is M ij Representing the risk field quality of a specific certain perception target and a perception label thereof; r is (r) i Representing the distance between the perception target and the own vehicle, v i Representing the corresponding speed of each perception target and the perception label, theta i Represents the included angle k between each perception target and the perception label and the driving speed direction of the own vehicle i For the undetermined coefficient, M (r i ,θ i ) Is defined in accordance with formula (2).
In summary, in the embodiment of the present application, by introducing the field of the driving wind field into the field of uncertainty planning, a tag probability risk field and a prediction dynamic risk field concept are provided on the basis of a traditional driving risk field theory, where the tag probability risk field is an artificial potential energy field for describing driving risks of a field when sensing ambiguity by taking into consideration a sensing tag probability distribution; the prediction dynamic risk field is an artificial potential energy field for describing the field driving risk with uncertainty of traffic vehicle position and speed prediction by utilizing a quality fuzzification idea; the two potential energy fields are respectively established for the perceived road obstacle at each moment, and are further integrated into a probability dynamic risk field of the whole moment, so that uncertain information is converted into definite information as far as possible, and a decision can be made on the basis of information blurring.
In step S103, a planned trace cluster is obtained by a gradient descent method in each preset time step based on the tag probability risk field and the predicted dynamic risk field, and post-processing is performed on the planned trace cluster to select an optimal path.
After the tag probability risk field and the predicted dynamic risk field are obtained, further, the embodiment of the application can also obtain correct path planning in various complex environments by obtaining the planning trace cluster and selecting the optimal track through post-processing.
Optionally, in an embodiment of the present application, based on the tag probability risk field and the predicted dynamic risk field, a planned trace cluster is obtained by a gradient descent method in each preset time step, and post-processing is performed on the planned trace cluster to select an optimal path, including: calculating repulsive force of the vehicle at the current moment, defining attractive force which always points to the target point position from the vehicle, obtaining resultant force of repulsive force and attractive force at the current moment through linear superposition, obtaining speed at the current moment, and obtaining planning speed at the next moment based on the speed at the current moment and the resultant force; iteratively calculating repulsive force, attractive force and planning speed at the vehicle position based on preset iteration times, and obtaining a plurality of vehicle position coordinates to form a planning track; continuously changing the repulsive force and the preset super-parameter of the attractive force to obtain candidate track clusters; and inputting the candidate track clusters into a preset evaluation function module, and calculating the score of each track, wherein the track corresponding to the highest score item is the optimal path.
Specifically, in each time step, the embodiment of the present application obtains a planned trace cluster by using a gradient descent method according to the probability dynamic risk field obtained at the moment, as shown in fig. 4, and the specific steps are as follows:
(1) And calculating the gradient of the risk field at the vehicle position at the current moment, and recording the gradient as repulsive force T. The gradient force is used to guide the vehicle away from the risk source. The attractive force F always pointing from the vehicle position to the target point position is defined, and its magnitude is defined as a certain constant. And the resultant force of repulsive force T and attractive force F at the moment is obtained by linear superposition and is recorded as N.
N＝c 1 T+c 2 F (4)
Wherein c 1 、c 2 Two self-defined parameters;
(2) Acquiring the current time speed v t The resultant force N and v are combined by using the principle of physical momentum t Obtaining the planning speed v at the next moment after linear superposition t+1 The specific superposition formula is as follows:
v t+1 ＝k 1 v t +k 2 N·Δt (5)
wherein k is 1 、k 2 For two coefficients to be determined, Δt is the size of each time step;
(3) According to the planning speed v obtained in step (2) t+1 Calculating the vehicle position after a time step, denoted as (x) 1 ，y 1 ) And is within (x) 1 ，y 1 ) Calculating the attractive force and repulsive force of the position to obtain resultant force, and then combining with v t+1 Linear superposition to obtain v t+2 ；
(4) Repeating the above steps n times to obtainSeries coordinates (x) 1 ，y 1 )、(x 2 ，y 2 )、(x 3 ，y 3 )……(x n ，y n ) A planned track is formed. Wherein n is a super parameter, and the specific size is set according to the calculation force limit;
(5) By constantly changing c 1 、c 2 The magnitudes of the two superparameters result in a series of different planned trajectories as candidate trajectory clusters, as shown in fig. 5.
Therefore, in the process of giving a planned path on the scene of the dynamic wind according to the existing probability, the embodiment of the application utilizes a gradient descent planning algorithm to give the field strength gradient force, the target point attractive force and the road field repulsive force at the corresponding positions at each moment, calculates the magnitude and the direction of the resultant force, and gives the plan of the subsequent moment according to the resultant force and the current speed, so that the path searching range is enlarged, and the decision robustness and universality are improved.
Further, the embodiment of the present application may input the obtained candidate track cluster as an input amount to the evaluation function module in the embodiment of the present application, so as to calculate the score of each track, and output the track with the highest score as the optimal path, where the evaluation indexes considered in the evaluation function are shown in table 1:
TABLE 1
As shown in table 1, the path length, the sum of squares of curvature derivatives, and the distance between the target point and the reference target point are used as path static evaluation indexes, the sum of squares of acceleration, the sum of squares of inverse acceleration, the maximum lateral acceleration, and the total time are used as path dynamic evaluation indexes, and then the eight evaluation indexes are weighted, superimposed and fitted to obtain an evaluation function, as shown in equation 6:
Wherein k is i To measure the weight coefficient of each index weight, C i Results obtained by the above index calculation methods are obtained.
It can be understood that the embodiment of the application is based on the "fuzzification" algorithm idea, performs fuzzification processing on uncertain classification probability information of the jump of the sensing end, distributes the quality points into a quality block, and all possible probability classifications are all linearly overlapped, and meanwhile, the original path planning speed and direction are added with a standard deviation as sigma at the prediction end γ ，σ θ The blurring process with the mean value being the original value presents a biased center mass risk field similar to a mass cone, thereby better adapting to the uncertainty of the probability dynamic risk field.
In order to further verify the reliability of the vehicle path planning method based on the risk field and the uncertain analysis, in the embodiment of the application, simulation experiments are carried out on Carla, THICV Sicity scenario and a Carsim-Simulink joint simulation platform, and the experimental comparison of the algorithm and the existing other algorithms is carried out by adopting the international automatic driving field-recognized TTC index, wherein the experimental results are as follows:
the result of the Carsim-Simulink experiment shows that: compared with the traditional risk field planning algorithm, the method has the advantages that 4.82 seconds of improvement is obtained on TTC indexes, and compared with the method which does not consider uncertain path planning algorithm, 10.94 seconds of improvement is obtained, namely the method has the advantage that the safety of the path is greatly improved under various working conditions with uncertain factors;
The result of Carla simulation experiment shows that: the planning method can accurately and rapidly give out reasonable path planning under various strong uncertain conditions such as a lane changing condition, an intersection turning condition, a foggy weather night obstacle avoidance condition and the like, and has high robustness, wherein the algorithm planning track under the lane changing condition is shown in fig. 6;
in addition, the application realizes sand table real vehicle simulation by using a THICV Sicity scenario platform, and the simulation result shows that: the planning method can be migrated to an actual real vehicle to realize rapid and effective path planning, and has certain robustness.
Based on the experimental results, the embodiment of the application can realize path planning under different complex working conditions and with uncertainty of the sensing end.
According to the vehicle path planning method based on the risk field and the uncertain analysis, the label classification and the position speed prediction result of the perceived target are obtained by carrying out label classification and position speed prediction on the perceived target; based on label classification and position speed prediction results of a perception target, respectively establishing a label probability risk field and a prediction dynamic risk field for measuring uncertainty in label classification and position speed prediction; and acquiring a planning trace cluster by a gradient descent method in each preset time step based on the tag probability risk field and the predicted dynamic risk field, and performing post-processing on the planning trace cluster to select an optimal path. According to the method, uncertainty factors are measured through a travelling risk field theory, and then an optimal path is selected according to the established probability risk field by using a gradient descent method, so that uncertain information of a sensing end in a special environment is fully utilized, robustness of decision control is improved, and accurate decision can be still carried out under the condition that the information of the sensing end has uncertain characteristics.
A vehicle path planning apparatus based on risk field and uncertainty analysis according to an embodiment of the present application will be described next with reference to the accompanying drawings.
Fig. 7 is a block schematic diagram of a vehicle path planning apparatus based on risk fields and uncertainty analysis in accordance with an embodiment of the present application.
As shown in fig. 7, the risk field and uncertainty analysis-based vehicle path planning apparatus 10 includes: prediction module 100, construction module 200, and acquisition module 300.
The prediction module 100 is configured to perform label classification and position velocity prediction on the perceived target, so as to obtain label classification and position velocity prediction results of the perceived target.
The construction module 200 is configured to respectively establish a tag probability risk field and a predicted dynamic risk field for measuring uncertainty in tag classification and position velocity prediction based on the tag classification and position velocity prediction results of the perceived target.
The selection module 300 is configured to obtain a planned trace cluster by a gradient descent method in each preset time step based on the tag probability risk field and the predicted dynamic risk field, and perform post-processing on the planned trace cluster to select an optimal path.
Optionally, in one embodiment of the present application, the prediction module 100 includes: training unit, removing unit, collecting unit, extracting unit and generating unit.
The training unit is used for constructing a multi-layer three-channel convolutional neural network and training the multi-layer three-channel convolutional neural network.
The removing unit is used for removing the last layer of the trained multi-layer three-channel convolutional neural network based on preset conditions and outputting the probability that the perception target belongs to each label.
And the collecting unit is used for extracting the time information of the perception target by using the long-period memory network encoder and collecting the long-period memory network states of all surrounding intelligent agents into one tensor in the social pooling layer.
And the extraction unit is used for extracting the space information of the perception target from the convolutional neural network.
The generation unit is used for generating probability distribution of the perception target in the transverse working condition and the longitudinal working condition by utilizing a preset number of long-short-term memory network decoders so as to obtain probability distribution prediction of the track and the speed of the perception target.
Optionally, in one embodiment of the present application, the building module 200 includes: the first measuring unit, the second measuring unit and the third measuring unit.
The first measuring unit is used for measuring the risk of the obstacle based on the tag probability risk field when the tag information meets the preset condition.
And the second measuring unit is used for measuring the target risk size based on the predicted dynamic risk field when the target speed information meets the preset condition.
And the third measuring unit is used for establishing a probability dynamic risk field based on the tag probability risk field and the prediction dynamic risk field so as to measure the risk size when the tag classification and the position speed prediction are uncertain.
Optionally, in one embodiment of the present application, the selecting module 300 includes: a first calculation unit, a second calculation unit, a change unit, and a third calculation unit.
The first calculation unit is used for calculating repulsive force at the vehicle position at the current moment, defining attractive force which always points to the target point position from the vehicle position, obtaining resultant force of the repulsive force and the attractive force at the current moment through linear superposition, obtaining the speed at the current moment, and obtaining planning speed at the next moment based on the speed at the current moment and the resultant force.
The second calculation unit is used for iteratively calculating repulsive force, attractive force and planning speed at the vehicle position based on the preset iteration times, and obtaining a plurality of vehicle position coordinates to form a planning track.
And the changing unit is used for continuously changing the preset super-parameter sizes of the repulsive force and the attractive force to obtain candidate track clusters.
And the third calculation unit is used for inputting the candidate track clusters into a preset evaluation function module, calculating the score of each track, and taking the track corresponding to the highest score item as the optimal path.
Optionally, in one embodiment of the present application, the calculation formula of the probabilistic dynamic risk field is as follows:
wherein,, V (V) i The risk field of the perception target appearing in the vision of all the vehicles, i is the number of the perception target and P ij For each probability of each label of each perceived target, j is the label number, M (r i ,θ i ) The risk field quality of the target and the target label thereof is perceived; r is (r) i To sense the distance between the target and the own vehicle, v i For each perceptionCorresponding speed of object and label, θ i For each included angle formed by the perception target, the target label and the running speed direction of the own vehicle, k i Is a coefficient to be determined.
It should be noted that the foregoing explanation of the embodiment of the vehicle path planning method based on the risk field and the uncertainty analysis is also applicable to the vehicle path planning device based on the risk field and the uncertainty analysis of the embodiment, and will not be repeated herein.
According to the vehicle path planning device based on the risk field and the uncertain analysis, the label classification and the position speed prediction result of the perceived target are obtained by carrying out label classification and position speed prediction on the perceived target; based on label classification and position speed prediction results of a perception target, respectively establishing a label probability risk field and a prediction dynamic risk field for measuring uncertainty in label classification and position speed prediction; and acquiring a planning trace cluster by a gradient descent method in each preset time step based on the tag probability risk field and the predicted dynamic risk field, and performing post-processing on the planning trace cluster to select an optimal path. According to the method, uncertainty factors are measured through a travelling risk field theory, and then an optimal path is selected according to the established probability risk field by using a gradient descent method, so that uncertain information of a sensing end in a special environment is fully utilized, robustness of decision control is improved, and accurate decision can be still carried out under the condition that the information of the sensing end has uncertain characteristics.
Fig. 8 is a schematic structural diagram of an electronic device according to an embodiment of the present application. The electronic device may include:
a memory 801, a processor 802, and a computer program stored on the memory 801 and executable on the processor 802.
The processor 802, when executing the program, implements the vehicle path planning method based on the risk field and uncertainty analysis provided in the above-described embodiment.
Further, the electronic device further includes:
a communication interface 803 for communication between the memory 801 and the processor 802.
A memory 801 for storing a computer program executable on the processor 802.
The memory 801 may include high-speed RAM memory or may further include non-volatile memory (non-volatile memory), such as at least one magnetic disk memory.
If the memory 801, the processor 802, and the communication interface 803 are implemented independently, the communication interface 803, the memory 801, and the processor 802 may be connected to each other through a bus and perform communication with each other. The bus may be an industry standard architecture (Industry Standard Architecture, abbreviated ISA) bus, an external device interconnect (Peripheral Component, abbreviated PCI) bus, or an extended industry standard architecture (Extended Industry Standard Architecture, abbreviated EISA) bus, among others. The buses may be divided into address buses, data buses, control buses, etc. For ease of illustration, only one thick line is shown in fig. 8, but not only one bus or one type of bus.
Alternatively, in a specific implementation, if the memory 801, the processor 802, and the communication interface 803 are integrated on a chip, the memory 801, the processor 802, and the communication interface 803 may communicate with each other through internal interfaces.
The processor 802 may be a central processing unit (Central Processing Unit, abbreviated as CPU), or an application specific integrated circuit (Application Specific Integrated Circuit, abbreviated as ASIC), or one or more integrated circuits configured to implement embodiments of the present application.
Embodiments of the present application also provide a computer readable storage medium having stored thereon a computer program which, when executed by a processor, implements a vehicle path planning method based on a risk field and uncertainty analysis as described above.
In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present application. In this specification, schematic representations of the above terms are not necessarily directed to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or N embodiments or examples. Furthermore, the different embodiments or examples described in this specification and the features of the different embodiments or examples may be combined and combined by those skilled in the art without contradiction.
Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present application, the meaning of "N" is at least two, such as two, three, etc., unless explicitly defined otherwise.
Any process or method descriptions in flow charts or otherwise described herein may be understood as representing modules, segments, or portions of code which include one or more executable instructions for implementing specific logical functions or steps of the process, and additional implementations are included within the scope of the preferred embodiment of the present application in which functions may be executed out of order from that shown or discussed, including substantially concurrently or in reverse order from that shown or discussed, depending on the functionality involved, as would be understood by those reasonably skilled in the art of the embodiments of the present application.
Logic and/or steps represented in the flowcharts or otherwise described herein, e.g., a ordered listing of executable instructions for implementing logical functions, can be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium would include the following: an electrical connection (electronic device) having one or N wires, a portable computer cartridge (magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program is printed, as the program may be electronically captured, via optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner if necessary, and then stored in a computer memory.
It is to be understood that portions of the present application may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the N steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented in hardware, as in another embodiment, may be implemented using any one or combination of the following techniques, as known in the art: discrete logic circuits having logic gates for implementing logic functions on data signals, application specific integrated circuits having suitable combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.
Those of ordinary skill in the art will appreciate that all or a portion of the steps carried out in the method of the above-described embodiments may be implemented by a program to instruct related hardware, where the program may be stored in a computer readable storage medium, and where the program, when executed, includes one or a combination of the steps of the method embodiments.
In addition, each functional unit in each embodiment of the present application may be integrated in one processing module, or each unit may exist alone physically, or two or more units may be integrated in one module. The integrated modules may be implemented in hardware or in software functional modules. The integrated modules may also be stored in a computer readable storage medium if implemented in the form of software functional modules and sold or used as a stand-alone product.
The above-mentioned storage medium may be a read-only memory, a magnetic disk or an optical disk, or the like. Although embodiments of the present application have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the application, and that variations, modifications, alternatives, and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the application.
Claims (12)
1. A vehicle path planning method based on risk field and uncertainty analysis, comprising the steps of:
performing label classification and position speed prediction on the perceived target to obtain label classification and position speed prediction results of the perceived target;
based on the label classification and position speed prediction results of the perception target, respectively establishing a label probability risk field and a prediction dynamic risk field for measuring uncertainty in the label classification and position speed prediction; and
and acquiring a planning trace cluster by a gradient descent method in each preset time step based on the tag probability risk field and the predicted dynamic risk field, and performing post-processing on the planning trace cluster to select an optimal path.
2. The method according to claim 1, wherein the performing label classification and position velocity prediction on the perceived target to obtain label classification and position velocity prediction results of the perceived target includes:
Constructing a multi-layer three-channel convolutional neural network, and training the multi-layer three-channel convolutional neural network;
based on preset conditions, removing the last layer of the trained multi-layer three-channel convolutional neural network, and outputting the probability that the perception target belongs to each label;
extracting time information of the perception target by using a long-period memory network encoder, and collecting long-period memory network states of all surrounding intelligent agents in a tensor in a social pooling layer;
extracting the space information of the perception target from the convolutional neural network;
and generating probability distribution of the perception target in the transverse working condition and the longitudinal working condition by using a preset number of long-short-term memory network decoders so as to obtain probability distribution prediction of the track and the speed of the perception target.
3. The method of claim 2, wherein the establishing a tag probability risk field and a predicted dynamic risk field that measure uncertainty in the tag classification and the location velocity prediction, respectively, based on the tag classification and the location velocity prediction of the perceived target, comprises:
when the tag information meets the preset condition, measuring the risk of the obstacle based on the tag probability risk field;
When the target speed information meets the preset condition, measuring the target risk size based on a predicted dynamic risk field;
and establishing a probability dynamic risk field based on the tag probability risk field and a prediction dynamic risk field to measure the risk when the tag classification and the position speed prediction are uncertain.
4. The method of claim 3, wherein the acquiring a planned trace cluster by a gradient descent method in each preset time step based on the tag probability risk field and the predicted dynamic risk field, and performing post-processing on the planned trace cluster to select an optimal path, comprises:
calculating repulsive force of a vehicle position at the current moment, defining attractive force which always points to a target point position from the vehicle position, obtaining resultant force of the repulsive force and the attractive force at the current moment through linear superposition, obtaining speed at the current moment, and obtaining planning speed at the next moment based on the speed at the current moment and the resultant force;
iteratively calculating repulsive force, attractive force and planning speed at the vehicle position based on preset iteration times, and obtaining a plurality of vehicle position coordinates to form a planning track;
continuously changing the preset super-parameter sizes of the repulsive force and the attractive force to obtain candidate track clusters;
And inputting the candidate track clusters into a preset evaluation function module, and calculating the score of each track, wherein the track corresponding to the highest score item is the optimal path.
5. A method according to claim 3, wherein the probabilistic dynamic risk field is calculated as:
wherein V is i For the risk fields of the perceived objects appearing in all the own vehicle vision, i is the number of the perceived object, P ij For each probability of each label of each perceived target, j is the label number, M (r i ,θ i ) The risk field quality of the target and the target label thereof is perceived; r is (r) i To sense the distance between the target and the own vehicle, v i For each perceived target and label corresponding speed, θ i For each included angle formed by the perception target, the target label and the running speed direction of the own vehicle, k i Is a coefficient to be determined.
6. A vehicle path planning apparatus based on a risk field and uncertainty analysis, comprising:
the prediction module is used for carrying out label classification and position speed prediction on the perception target to obtain label classification and position speed prediction results of the perception target;
the construction module is used for respectively establishing a tag probability risk field and a prediction dynamic risk field for measuring uncertainty in tag classification and position speed prediction based on the tag classification and position speed prediction result of the perception target; and
The selection module is used for acquiring a planning trace cluster through a gradient descent method in each preset time step based on the tag probability risk field and the prediction dynamic risk field, and performing post-processing on the planning trace cluster to select an optimal path.
7. The apparatus of claim 6, wherein the prediction module comprises:
the training unit is used for constructing a multi-layer three-channel convolutional neural network and training the multi-layer three-channel convolutional neural network;
the removing unit is used for removing the last layer of the trained multi-layer three-channel convolutional neural network based on preset conditions and outputting the probability that the perception target belongs to each label;
the collecting unit is used for extracting the time information of the perception target by utilizing the long-period memory network encoder and collecting the long-period memory network states of all surrounding intelligent agents into a tensor in a social pooling layer;
the extraction unit is used for extracting the space information of the perception target through a convolutional neural network;
the generation unit is used for generating probability distribution of the perception target in the transverse working condition and the longitudinal working condition by utilizing the preset number of long-short-term memory network decoders so as to obtain probability distribution prediction of the track and the speed of the perception target.
8. The apparatus of claim 7, wherein the build module comprises:
the first measuring unit is used for measuring the risk of the obstacle based on the tag probability risk field when the tag information meets the preset condition;
the second measuring unit is used for measuring the target risk size based on the predicted dynamic risk field when the target position and speed information meets the preset condition;
and the third measuring unit is used for establishing a probability dynamic risk field based on the tag probability risk field and the prediction dynamic risk field so as to measure the risk when the tag classification and the position speed prediction are uncertain.
9. The apparatus of claim 8, wherein the means for selecting comprises:
the first calculation unit is used for calculating repulsive force at the vehicle position at the current moment, defining attractive force which is always pointed to the target point position by the vehicle position, obtaining resultant force of the repulsive force and the attractive force at the current moment through linear superposition, obtaining the speed at the current moment, and obtaining planning speed at the next moment based on the speed at the current moment and the resultant force;
the second calculation unit is used for iteratively calculating repulsive force, attractive force and planning speed at the vehicle position based on preset iteration times, and obtaining a plurality of vehicle position coordinates to form a planning track;
A changing unit, configured to continuously change the repulsive force and the preset super-parameter of the attractive force to obtain a candidate track cluster;
and the third calculation unit is used for inputting the candidate track clusters into a preset evaluation function module, calculating the score of each track, and taking the track corresponding to the highest score item as the optimal path.
10. The apparatus of claim 8, wherein the probabilistic dynamic risk field is calculated as:
wherein V is i For the risk fields of the perceived objects appearing in all the own vehicle vision, i is the number of the perceived object, P ij For each probability of each label of each perceived target, j is the label number, M (r i ,θ i ) The risk field quality of the target and the target label thereof is perceived; r is (r) i To sense the distance between the target and the own vehicle, v i For each perceived target and label corresponding speed, θ i For each perceived target, targetThe included angle k formed by the label and the running speed direction of the bicycle i Is a coefficient to be determined.
11. A vehicle, characterized by comprising: a memory, a processor and a computer program stored on the memory and executable on the processor, the processor executing the program to implement the risk-field and uncertainty analysis based vehicle path planning method of any of claims 1-5.
12. A computer readable storage medium having stored thereon a computer program, the program being executable by a processor for implementing a risk-field and uncertainty analysis based vehicle path planning method according to any of claims 1-5.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310323365.7A CN116448134B (en) | 2023-03-29 | 2023-03-29 | Vehicle path planning method and device based on risk field and uncertain analysis |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310323365.7A CN116448134B (en) | 2023-03-29 | 2023-03-29 | Vehicle path planning method and device based on risk field and uncertain analysis |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN116448134A true CN116448134A (en) | 2023-07-18 |
| CN116448134B CN116448134B (en) | 2023-11-07 |
Family
ID=87124864
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310323365.7A Active CN116448134B (en) | 2023-03-29 | 2023-03-29 | Vehicle path planning method and device based on risk field and uncertain analysis |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116448134B (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117961915A (en) * | 2024-03-28 | 2024-05-03 | 太原理工大学 | Intelligent auxiliary decision-making method of coal mine tunneling robot |
Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN113312924A (en) * | 2021-06-23 | 2021-08-27 | 北京鼎泰智源科技有限公司 | Risk rule classification method and device based on NLP high-precision analysis label |
| EP3971526A1 (en) * | 2020-09-18 | 2022-03-23 | Zenuity AB | Path planning in autonomous driving environments |
| CN114543827A (en) * | 2022-02-11 | 2022-05-27 | 齐鲁工业大学 | Path planning method and device |
| CN114610016A (en) * | 2022-01-25 | 2022-06-10 | 合肥工业大学 | Intelligent vehicle collision avoidance path planning method based on dynamic virtual expansion of barrier |
| CN114894206A (en) * | 2022-03-18 | 2022-08-12 | 北京智行者科技有限公司 | Path planning method and device, vehicle and storage medium |
| CN115167463A (en) * | 2022-08-09 | 2022-10-11 | 广西大学 | Path planning method based on combination of neural network and artificial potential field method |
| CN115683145A (en) * | 2022-11-03 | 2023-02-03 | 北京踏歌智行科技有限公司 | Automatic driving safety obstacle avoidance method based on track prediction |
-
2023
- 2023-03-29 CN CN202310323365.7A patent/CN116448134B/en active Active
Patent Citations (7)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| EP3971526A1 (en) * | 2020-09-18 | 2022-03-23 | Zenuity AB | Path planning in autonomous driving environments |
| CN113312924A (en) * | 2021-06-23 | 2021-08-27 | 北京鼎泰智源科技有限公司 | Risk rule classification method and device based on NLP high-precision analysis label |
| CN114610016A (en) * | 2022-01-25 | 2022-06-10 | 合肥工业大学 | Intelligent vehicle collision avoidance path planning method based on dynamic virtual expansion of barrier |
| CN114543827A (en) * | 2022-02-11 | 2022-05-27 | 齐鲁工业大学 | Path planning method and device |
| CN114894206A (en) * | 2022-03-18 | 2022-08-12 | 北京智行者科技有限公司 | Path planning method and device, vehicle and storage medium |
| CN115167463A (en) * | 2022-08-09 | 2022-10-11 | 广西大学 | Path planning method based on combination of neural network and artificial potential field method |
| CN115683145A (en) * | 2022-11-03 | 2023-02-03 | 北京踏歌智行科技有限公司 | Automatic driving safety obstacle avoidance method based on track prediction |
Non-Patent Citations (1)
| Title |
|---|
| 王建强;吴剑;李洋;: "基于人-车-路协同的行车风险场概念、原理及建模", 中国公路学报, no. 01 * |
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117961915A (en) * | 2024-03-28 | 2024-05-03 | 太原理工大学 | Intelligent auxiliary decision-making method of coal mine tunneling robot |
| CN117961915B (en) * | 2024-03-28 | 2024-06-04 | 太原理工大学 | Intelligent auxiliary decision-making method of coal mine tunneling robot |
Also Published As
| Publication number | Publication date |
|---|---|
| CN116448134B (en) | 2023-11-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN111971574B (en) | Deep learning based feature extraction for LIDAR localization of autonomous vehicles | |
| CN111367282B (en) | Robot navigation method and system based on multimode perception and reinforcement learning | |
| CN111771135B (en) | LIDAR positioning using RNN and LSTM for time smoothing in autonomous vehicles | |
| CN112212874B (en) | Vehicle track prediction method and device, electronic equipment and computer readable medium | |
| CN112347993B (en) | Expressway vehicle behavior and track prediction method based on vehicle-unmanned aerial vehicle cooperation | |
| Hu et al. | Safe local motion planning with self-supervised freespace forecasting | |
| JP2022516383A (en) | Autonomous vehicle planning | |
| CN114384920A (en) | Dynamic obstacle avoidance method based on real-time construction of local grid map | |
| CN110389584A (en) | Method for assessing the track candidate item of automatic driving vehicle | |
| US11501449B2 (en) | Method for the assessment of possible trajectories | |
| Chen et al. | Driving maneuvers prediction based autonomous driving control by deep Monte Carlo tree search | |
| CN111771141A (en) | LIDAR positioning in autonomous vehicles using 3D CNN networks for solution inference | |
| CN107310550A (en) | Road vehicles travel control method and device | |
| Ivanovic et al. | Injecting planning-awareness into prediction and detection evaluation | |
| CN113128381A (en) | Obstacle trajectory prediction method, system and computer storage medium | |
| CN116448134B (en) | Vehicle path planning method and device based on risk field and uncertain analysis | |
| JP2023523350A (en) | Vehicle-based data processing method, data processing apparatus, computer apparatus, and computer program | |
| Camara et al. | A heuristic model for pedestrian intention estimation | |
| Sharma et al. | Kernelized convolutional transformer network based driver behavior estimation for conflict resolution at unsignalized roundabout | |
| Dong et al. | Interactive trajectory prediction for autonomous driving via recurrent meta induction neural network | |
| Khalil et al. | Exploiting multi-modal fusion for urban autonomous driving using latent deep reinforcement learning | |
| Kawasaki et al. | Multimodal trajectory predictions for autonomous driving without a detailed prior map | |
| CN115062202A (en) | Method, device, equipment and storage medium for predicting driving behavior intention and track | |
| Masmoudi et al. | Autonomous car-following approach based on real-time video frames processing | |
| Srinivasan et al. | Beyond RMSE: Do machine-learned models of road user interaction produce human-like behavior? |
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 | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |