Long time horizon driving behavior decision method suitable for high speed and loop traffic scenarios

Abstract

The application relates to the technical field of vehicles, in particular to a long-time-domain driving behavior decision method suitable for high-speed and loop traffic scenes, which comprises the following steps: acquiring a global planning path of a vehicle, a current motion state of the vehicle and current motion states of all surrounding vehicles in a region; generating an optimal driving behavior sequence of the vehicle according to the global planning path, the current motion state of the vehicle and the current motion states of all surrounding vehicles in the region; planning a driving track of the vehicle based on a first driving behavior of the optimal driving behavior sequence, and regenerating the optimal driving behavior sequence after the vehicle executes the first driving behavior based on the driving track, until the global planning path is completed. Therefore, the embodiment of the application can realize multi-step decision of driving behaviors such as lane-changing gap selection and lane-changing overtaking, has foresight, and simultaneously gives an optimal solution to long-time-domain driving behavior sequence planning based on feasibility discrimination, considers safety and high efficiency, and meets the needs of general driving.

Description

Technical Field

[0001] This application relates to the field of vehicle technology, and in particular to a long-term driving behavior decision-making method applicable to highway and ring road traffic scenarios. Background Technology

[0002] Driving decision-making is generally divided into global path planning (lane level), driving behavior planning (semantic driving behaviors such as lane changing and avoidance), and motion trajectory planning (the specific motion trajectory of the target driving behavior).

[0003] Traditional behavior planning methods primarily rely on single-step planning, which has a short-sighted drawback and cannot plan multiple behavior sequences to achieve optimal driving over a longer period. Taking highway scenarios as an example, the challenge of long-term behavior decision-making lies in integrating predictions of surrounding vehicle interactions during the decision-making process and achieving real-time solutions to complex decision problems. Summary of the Invention

[0004] This application provides a long-term driving behavior decision-making method, device, vehicle, and storage medium applicable to highway and ring road traffic scenarios. It can realize multi-step decision-making for driving behaviors such as lane change interval selection and lane changing overtaking, and has forward-looking characteristics. At the same time, based on feasibility judgment, it provides the optimal solution for long-term driving behavior sequence planning, taking into account both safety and efficiency, and meeting the needs of general driving.

[0005] The first aspect of this application provides a long-term driving behavior decision-making method applicable to highway and ring road traffic scenarios, comprising the following steps: obtaining the global planning path of a vehicle, the current motion state of the vehicle, and the current motion states of all surrounding vehicles in the area; generating an optimal driving behavior sequence for the vehicle based on the global planning path, the current motion state of the vehicle, and the current motion states of all surrounding vehicles in the area; planning the driving trajectory of the vehicle based on the first driving behavior of the optimal driving behavior sequence, and after controlling the vehicle to execute the first driving behavior based on the driving trajectory, regenerating the optimal driving behavior sequence until the global planning path is completed.

[0006] Optionally, generating the optimal driving behavior sequence for the vehicle based on the global planning path, the current motion state of the vehicle, and the current motion states of all surrounding vehicles in the area includes: using the current motion state of the vehicle and the current motion states of all surrounding vehicles in the area as the root node, sequentially selecting the optimal child nodes until a leaf node is reached; at the leaf node, calculating the new child node state of the vehicle under all driving behaviors, and performing reachability calculations on each new child node state based on the motion model of the vehicle and the motion models of all surrounding vehicles in the area to obtain reachable nodes; randomly selecting a child node as an extension node among all reachable child nodes, and starting from the extension node, reaching the target state based on the Rollout strategy to obtain simulation results; performing backpropagation based on the simulation results, updating the evaluation values ​​of all nodes on the child node path until the iteration stopping condition is met, determining the optimal path based on the evaluation values ​​of all nodes, and generating the optimal driving behavior sequence based on the driving behaviors corresponding to the nodes on the optimal path.

[0007] Optionally, each node stores the motion state of the vehicle and all surrounding vehicles within the area, and updates the motion state of the vehicle and all surrounding vehicles within the area in each node based on the motion model of the vehicle and the motion model of all surrounding vehicles within the area.

[0008] Optionally, based on the vehicle's motion model and the motion models of all surrounding vehicles in the area, reachability node calculation is performed on the state of each new child node to obtain reachable nodes. This includes: predicting the lateral and longitudinal behaviors of all surrounding vehicles in the area based on the current motion state of the vehicle and the current motion states of all surrounding vehicles in the area, to obtain lateral and longitudinal behavior prediction results; matching the motion models of all surrounding vehicles in the area based on the lateral and longitudinal behavior prediction results; verifying whether the state of each new child node satisfies the vehicle's behavior constraints based on the vehicle's motion model and the motion models of all surrounding vehicles in the area; and determining that the node is reachable when the vehicle's behavior constraints are satisfied.

[0009] Optionally, based on the current motion state of the vehicle and the current motion states of all surrounding vehicles in the area, the lateral and longitudinal behaviors of all surrounding vehicles in the area are predicted to obtain the lateral and longitudinal behavior prediction results. This includes: inputting the current motion state and the predicted current motion states of all surrounding vehicles in the area into a pre-trained dynamic Bayesian network for lane-changing behavior prediction, and outputting the lane-changing behavior confidence scores of all surrounding vehicles in the area. The variable information in the dynamic Bayesian network for lane-changing behavior prediction includes lane-changing area, lane-changing intention, lane-changing drive, lateral distance of lane lines, lateral speed, speed difference with the preceding vehicle, and longitudinal distance difference. The variable information in the dynamic Bayesian network for yielding behavior prediction includes speed conditions, distance conditions, yielding intention, longitudinal position of the vehicle, longitudinal speed of the vehicle, longitudinal position of the following vehicle, longitudinal speed of the following vehicle, and yielding / cutting-off completion indicators.

[0010] Optionally, in the motion model of the self-driving vehicle and the motion model of the surrounding vehicles, the longitudinal acceleration and the time to complete a sequence corresponding to each driving behavior are preset values; the motion model of the surrounding vehicles further includes: for longitudinal following behavior, setting the lane leader in the observation area to drive at a constant speed, and the following vehicles in the lane adopting an IDM model (Intelligence Driver Model); for longitudinal yielding behavior, generating a virtual leading vehicle in front of the self-driving vehicle, and controlling the longitudinal speed of the self-driving vehicle based on the IDM model, when there is a conflict between surrounding vehicles, the following vehicles yield, and when there is a conflict between surrounding vehicles and the self-driving vehicle, the yielding vehicle is determined to yield when the yielding confidence is greater than a first preset confidence level; for lateral lane changing behavior, when the lane-changing vehicle changes to the outermost lane and changes one lane at a time, setting the lane leader in the observation area to drive at a constant speed, and the following vehicles in the lane adopting an IDM model, when there is a conflict between the lane-changing vehicle and the original lane vehicle, the following vehicles yield, and the lane-changing vehicle is determined to have a lane-changing behavior when the lane-changing confidence is greater than a second preset confidence level.

[0011] Optionally, the behavioral constraints include collision detection, yield verification, drivable area verification, maximum speed verification, and endpoint position verification. The collision detection includes: after the vehicle's driving behavior ends, determining whether the distance and time distance between the vehicle and the vehicles in front and behind it in the lane are both greater than constraint values. If they are greater, the driving behavior is deemed feasible; otherwise, it is deemed infeasible. Here, non-vehicle obstacles are virtualized as surrounding vehicles with the same speed as the obstacles. The yield verification includes: when the vehicle performs a lane-changing behavior, if the yield confidence of a vehicle conflicting with the vehicle is less than a preset confidence level, the lane-changing behavior is deemed infeasible; otherwise, it is deemed feasible. The drivable area... The domain verification includes: after the driving behavior of the vehicle ends, if the vehicle's position is not within the drivable area, the driving behavior is determined to be infeasible; otherwise, the driving behavior is determined to be feasible. The maximum speed verification includes: at the start of the vehicle's acceleration behavior, if the vehicle's speed is greater than or equal to the speed limit of the current position, acceleration is determined to be infeasible; otherwise, after the acceleration behavior ends, if the vehicle's speed is greater than or equal to the speed limit of the current position, the maximum speed of the vehicle is determined to be the speed limit. The endpoint position verification includes: after the driving behavior of the vehicle ends, if the vehicle's longitudinal position exceeds the target position and is not within the target lane, the driving behavior is determined to be infeasible.

[0012] Optionally, the driving behavior includes any one of accelerating straight, driving straight at a constant speed, decelerating straight, changing lanes to the left, or changing lanes to the right.

[0013] A second aspect of this application provides a long-term driving behavior decision-making device suitable for highway and ring road traffic scenarios, comprising: an information acquisition module for acquiring the global planned path of a vehicle, the current motion state of the vehicle, and the current motion states of all surrounding vehicles within the area; a decision-making module for generating an optimal driving behavior sequence for the vehicle based on the global planned path, the current motion state of the vehicle, and the current motion states of all surrounding vehicles within the area; and a control module for planning the driving trajectory of the vehicle based on the first driving behavior of the optimal driving behavior sequence, and after controlling the vehicle to execute the first driving behavior based on the driving trajectory, regenerating the optimal driving behavior sequence until the global planned path is completed.

[0014] Optionally, the decision module is configured to: take the current motion state of the vehicle and the current motion states of all surrounding vehicles in the area as the root node, sequentially select the optimal child node until the leaf node; at the leaf node, calculate the new child node state of the vehicle under all driving behaviors, and perform reachability calculation on each new child node state based on the motion model of the vehicle and the motion model of all surrounding vehicles in the area to obtain reachable nodes; randomly select a child node as an extension node among all reachable child nodes, and start from the extension node to reach the target state based on the Rollout strategy to obtain simulation results; perform backpropagation based on the simulation results, update the evaluation values ​​of all nodes on the child node path until the iteration stopping condition is met, determine the optimal path based on the evaluation values ​​of all nodes, and generate the optimal driving behavior sequence based on the driving behaviors corresponding to the nodes on the optimal path.

[0015] Optionally, each node stores the motion state of the vehicle and all surrounding vehicles within the area, and updates the motion state of the vehicle and all surrounding vehicles within the area in each node based on the motion model of the vehicle and the motion model of all surrounding vehicles within the area.

[0016] Optionally, the decision module is further configured to: predict the lateral and longitudinal behaviors of all surrounding vehicles in the region based on the current motion state of the vehicle and the current motion state of all surrounding vehicles in the region, and obtain lateral and longitudinal behavior prediction results; match the motion models of all surrounding vehicles in the region based on the lateral and longitudinal behavior prediction results; verify whether the state of each new child node satisfies the behavior constraints of the vehicle based on the motion model of the vehicle and the motion models of all surrounding vehicles in the region; and determine that the node is reachable when the behavior constraints of the vehicle are satisfied.

[0017] Optionally, the decision module is further configured to: input the current motion state and the predicted current motion state of all surrounding vehicles in the area into a pre-trained dynamic Bayesian network for lane-changing behavior prediction, and output the confidence scores of lane-changing behavior for all surrounding vehicles in the area, wherein the variable information in the dynamic Bayesian network for lane-changing behavior prediction includes lane-changing zone, lane-changing intention, lane-changing drive, lateral distance of lane line, lateral speed, speed difference with the preceding vehicle, and longitudinal distance difference; input the current motion state and the predicted current motion state of all surrounding vehicles in the area into a pre-trained dynamic Bayesian network for yielding behavior prediction, and output the yielding confidence scores, wherein the variable information in the dynamic Bayesian network for yielding behavior prediction includes speed conditions, distance conditions, yielding intention, longitudinal position of the vehicle, longitudinal speed of the vehicle, longitudinal position of the following vehicle, longitudinal speed of the following vehicle, and yielding / cutting-off completion indicators.

[0018] Optionally, in the motion model of the vehicle and the motion model of the surrounding vehicles, the longitudinal acceleration and the time to complete a sequence corresponding to each driving behavior are preset values; the motion model of the surrounding vehicles further includes: for longitudinal following behavior, setting the lane leader in the observation area to drive at a constant speed, and the following vehicles in the lane adopting an IDM model; for longitudinal yielding behavior, generating a virtual leading vehicle in front of the vehicle, and controlling the longitudinal speed of the vehicle based on the IDM model, when there is a conflict between surrounding vehicles, the following vehicles yield, and when there is a conflict between surrounding vehicles and the vehicle, the surrounding vehicles are determined to yield when the yielding confidence is greater than a first preset confidence level; for lateral lane changing behavior, when the lane changing vehicle changes to the outermost lane and changes one lane at a time, setting the lane leader in the observation area to drive at a constant speed, and the following vehicles in the lane adopting an IDM model, when there is a conflict between the lane changing vehicle and the original lane vehicle, the following vehicles yield, and the lane changing confidence is determined to have lane changing behavior when the lane changing confidence is greater than a second preset confidence level.

[0019] Optionally, the behavioral constraints include collision detection, yield verification, drivable area verification, maximum speed verification, and endpoint position verification. The collision detection includes: after the vehicle's driving behavior ends, determining whether the distance and time distance between the vehicle and the vehicles in front and behind it in the lane are both greater than constraint values. If they are greater, the driving behavior is deemed feasible; otherwise, it is deemed infeasible. Here, non-vehicle obstacles are virtualized as surrounding vehicles with the same speed as the obstacles. The yield verification includes: when the vehicle performs a lane-changing behavior, if the yield confidence of a vehicle conflicting with the vehicle is less than a preset confidence level, the lane-changing behavior is deemed infeasible; otherwise, it is deemed feasible. The drivable area... The domain verification includes: after the driving behavior of the vehicle ends, if the vehicle's position is not within the drivable area, the driving behavior is determined to be infeasible; otherwise, the driving behavior is determined to be feasible. The maximum speed verification includes: at the start of the vehicle's acceleration behavior, if the vehicle's speed is greater than or equal to the speed limit of the current position, acceleration is determined to be infeasible; otherwise, after the acceleration behavior ends, if the vehicle's speed is greater than or equal to the speed limit of the current position, the maximum speed of the vehicle is determined to be the speed limit. The endpoint position verification includes: after the driving behavior of the vehicle ends, if the vehicle's longitudinal position exceeds the target position and is not within the target lane, the driving behavior is determined to be infeasible.

[0020] Optionally, the driving behavior includes any one of accelerating straight, driving straight at a constant speed, decelerating straight, changing lanes to the left, or changing lanes to the right.

[0021] A third aspect of this application provides a vehicle, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the long-term driving behavior decision-making method applicable to highway and ring road traffic scenarios as described in the above embodiments.

[0022] A fourth aspect of this application provides a computer-readable storage medium having a computer program stored thereon, which is executed by a processor to implement the long-term driving behavior decision-making method applicable to highway and ring road traffic scenarios as described in the above embodiments.

[0023] Therefore, this application has at least the following beneficial effects:

[0024] For multi-lane driving scenarios such as highways and ring roads, based on MCTS (Monte Carlo Tree Search), it can realize multi-step decision-making for driving behaviors such as lane change interval selection and lane overtaking by coupling the optimal decision objective in highway scenarios with the prediction of surrounding vehicle behavior response, which has a forward-looking nature. At the same time, based on feasibility judgment, it provides the optimal solution for long-term driving behavior sequence planning, taking into account both safety and efficiency, and meeting the needs of general driving.

[0025] Additional aspects and advantages of this 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 this application. Attached Figure Description

[0026] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0027] Figure 1 This is a flowchart of a long-term driving behavior decision-making method applicable to highway and ring road traffic scenarios, according to an embodiment of this application.

[0028] Figure 2 A schematic diagram of the Freina coordinate system for a highway lane, according to an embodiment of this application;

[0029] Figure 3 The overall workflow diagram of the decision-making system is based on an embodiment of this application.

[0030] Figure 4 The embodiment of this application is a functional diagram of the behavior decision algorithm;

[0031] Figure 5 A schematic diagram of the behavioral decision-making process according to an embodiment of this application;

[0032] Figure 6A schematic diagram of a dynamic Bayesian network for lane-changing behavior prediction according to an embodiment of this application;

[0033] Figure 7 A schematic diagram of a dynamic Bayesian network for predicting yielding behavior according to an embodiment of this application;

[0034] Figure 8 This is an example diagram of a long-term driving behavior decision-making device applicable to highway and ring road traffic scenarios according to an embodiment of this application;

[0035] Figure 9 This is a structural schematic diagram of a vehicle according to an embodiment of this application. Detailed Implementation

[0036] The embodiments of this application are described in detail below. Examples of these embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals denote the same or similar elements or elements having the same or similar functions throughout. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain this application, and should not be construed as limiting this application.

[0037] The following description, with reference to the accompanying drawings, outlines a long-time-domain driving behavior decision-making method, apparatus, vehicle, and storage medium applicable to highway and ring road traffic scenarios. Addressing the problems mentioned in the background section, this application provides a long-time-domain driving behavior decision-making method applicable to highway and ring road traffic scenarios. Specifically, Figure 1 This is a flowchart illustrating a long-term driving behavior decision-making method applicable to highway and ring road traffic scenarios, provided as an embodiment of this application.

[0038] like Figure 1 As shown, this long-term driving behavior decision-making method applicable to highway and ring road traffic scenarios includes the following steps:

[0039] In step S101, the global planned path of the vehicle, the current motion state of the vehicle, and the current motion state of all surrounding vehicles in the area are obtained.

[0040] In this application embodiment, the global planned path of the vehicle, the current motion state of the vehicle, and the current motion state of all surrounding vehicles in the area can be obtained in various ways, without specific limitations.

[0041] It should be noted that, in the embodiments of this application, the above road and vehicle position information can be converted from the Cartesian coordinate system to the FRINA coordinate system based on the reference lane centerline. Specifically, the FRINA coordinate system for highway lanes can be as follows: Figure 2 As shown, the position coordinates of a point in the Flyner coordinate system can be expressed as:

[0042] (Longitudinal displacement s, to reference line displacement d).

[0043] In step S102, the optimal driving behavior sequence of the vehicle is generated based on the global planning path, the current motion state of the vehicle, and the current motion state of all surrounding vehicles in the area.

[0044] Driving behavior can include any one of the following: accelerating straight ahead, driving straight ahead at a constant speed, decelerating straight ahead, changing lanes to the left, or changing lanes to the right.

[0045] It is understood that the embodiments of this application can calculate the target lane and longitudinal distance to be reached after a vehicle takes 6 steps based on high-precision maps, global path planning, and average vehicle speed, in order to generate an optimal driving behavior sequence. For example, the embodiments of this application can calculate the target lane and longitudinal distance to be reached after 6 steps, and each step can take a fixed duration, such as 4 seconds.

[0046] Specifically, such as Figure 3 As shown, in this embodiment of the application, a high-precision map, global path planning on the map, the surrounding vehicle motion state and the driver's motion state within the region can be used as upper-level inputs. The optimal driving behavior sequence can be planned through a decision-maker. The decision-maker can be as follows: Figure 4 As shown, based on the global traffic information within the input area (high-precision road map, surrounding vehicle movement status, global path planning of the vehicle, road speed limit, traffic control and other information), a driving behavior sequence such as merging left → accelerating straight → driving straight at a constant speed → merging right → driving straight at a constant speed can be planned.

[0047] In this embodiment, the optimal driving behavior sequence of the vehicle is generated based on the global planning path, the current motion state of the vehicle, and the current motion states of all surrounding vehicles in the area. This includes: using the current motion state of the vehicle and the current motion states of all surrounding vehicles in the area as the root node, sequentially selecting the optimal child nodes until a leaf node is reached; at the leaf node, calculating the new child node state of the vehicle under all driving behaviors, and performing reachability calculations on each new child node state based on the vehicle's motion model and the motion models of all surrounding vehicles in the area to obtain reachable nodes; randomly selecting a child node as an extension node from all reachable child nodes, and starting from the extension node, reaching the target state based on the Rollout strategy to obtain simulation results; performing backpropagation based on the simulation results to update the evaluation values ​​of all nodes on the child node path until the iteration stopping condition is met, and determining the optimal path based on the evaluation values ​​of all nodes, and generating the optimal driving behavior sequence based on the driving behaviors corresponding to the nodes on the optimal path.

[0048] It is understood that embodiments of this application may provide an algorithm for generating optimal driving sequences, so as to generate optimal driving behavior sequences using this algorithm. Specifically, as... Figure 5As shown in the figure, the long-term driving behavior decision-making algorithm applicable to highway and ring road traffic scenarios in this application embodiment is as follows:

[0049] Input: High-precision map (including global target path), regional vehicle movement status, and vehicle movement status;

[0050] Output: The optimal sequence of driving behaviors.

[0051] 1: Initialization. The root node is the initial state of the self-car and the surrounding car.

[0052] 2: Selection. Starting from the root node, select the best child node in sequence based on the evaluation value, until the leaf node is reached.

[0053] 3: Reachable Node Calculation. Termination condition not met: At this leaf node, calculate the new child node state for all autonomous vehicle behaviors. The autonomous vehicle and surrounding vehicle state changes are calculated based on the proposed motion model. Reachability verification is performed to filter out feasible driving behaviors and their corresponding reachable nodes.

[0054] Termination condition met: Stop the loop and proceed to step 7.

[0055] 4: Simulation. Randomly select one of the reachable child nodes as the expansion node for this round. Starting from this node, reach the target state based on the Rollout strategy.

[0056] 5: Backpropagation. Based on the simulation results, update the evaluation values ​​of all nodes on the child node path.

[0057] 6: Repeat steps 2 through 5.

[0058] 7: Return: The sequence of vehicle behavior corresponding to the path selected by the child node.

[0059] In this embodiment of the application, each node stores the motion state of the vehicle and all surrounding vehicles in the area, and updates the motion state of the vehicle and all surrounding vehicles in the area in each node based on the motion model of the vehicle and the motion model of all surrounding vehicles in the area.

[0060] Understandably, in MCTS, each node stores the motion states of all vehicles in that step (including the vehicle itself and surrounding vehicles), its connections with preceding and following nodes (i.e., the driving behaviors taken by the vehicle), the number of times the node has been visited, and its evaluation function value. Based on a certain node, after the vehicle takes a certain behavior and experiences a step, the vehicle motion states in the generated node need to be updated. Specifically, to store the motion states and positional relationships of all vehicles within the storage area, this embodiment can establish a linked list for each lane; in the linked list, each cell stores the corresponding vehicle motion state (longitudinal position, longitudinal speed), and its preceding and following adjacent cells (i.e., the relationship between preceding and following vehicles). When a vehicle changes lanes (including the vehicle itself and surrounding vehicles), the conflict relationship is first determined based on the longitudinal position, and the preceding and following positional relationship with the conflicting vehicle after the lane change is determined. Then, the linked list of the involved lanes is updated based on the preceding and following connections of the linked list.

[0061] In this embodiment, reachability node calculation is performed on the state of each new sub-node based on the vehicle's motion model and the motion models of all surrounding vehicles in the area to obtain reachable nodes. This includes: predicting the lateral and longitudinal behaviors of all surrounding vehicles in the area based on the current motion state of the vehicle and the current motion states of all surrounding vehicles in the area to obtain lateral and longitudinal behavior prediction results; matching the motion models of all surrounding vehicles in the area based on the lateral and longitudinal behavior prediction results; verifying whether the state of each new sub-node satisfies the vehicle's behavior constraints based on the vehicle's motion model and the motion models of all surrounding vehicles in the area; and determining that the node is reachable when the vehicle's behavior constraints are satisfied.

[0062] It is understood that the embodiments of this application can design motion models for the vehicle and the surrounding vehicles based on the designed lateral-longitudinal behavior prediction; and design behavior feasibility verification based on the above motion models. The following will elaborate on the lateral-longitudinal behavior prediction of the surrounding vehicles, the design of the motion models for the vehicle and the surrounding vehicles, and the feasibility verification, as detailed below:

[0063] In this embodiment, the lateral and longitudinal behaviors of all surrounding vehicles in the area are predicted based on the current motion state of the vehicle and the current motion states of all surrounding vehicles in the area, resulting in lateral and longitudinal behavior predictions. This includes: inputting the current motion state and the predicted current motion states of all surrounding vehicles in the area into a pre-trained dynamic Bayesian network for lane-changing behavior prediction, and outputting the confidence scores of lane-changing behavior for all surrounding vehicles in the area. The variable information in the dynamic Bayesian network for lane-changing behavior prediction includes lane-changing area, lane-changing intention, lane-changing drive, lateral distance of lane lines, lateral speed, speed difference with the preceding vehicle, and longitudinal distance difference; and inputting the current motion state and the predicted current motion states of all surrounding vehicles in the area into a pre-trained dynamic Bayesian network for yielding behavior prediction, and outputting yielding confidence scores. The variable information in the dynamic Bayesian network for yielding behavior prediction includes speed conditions, distance conditions, yielding intention, longitudinal position of the vehicle, longitudinal speed of the vehicle, longitudinal position of the following vehicle, longitudinal speed of the following vehicle, and yielding / cutting completion indicators.

[0064] It is understood that the embodiments of this application can output the confidence level of lane-changing behavior of surrounding vehicles and the confidence level of yielding behavior when going straight, based on dynamic Bayesian networks. Here, the confidence level can refer to the degree to which the true value of a certain population parameter of a sample has a certain probability of falling around the measurement result.

[0065] Specifically, the specific steps for predicting the lateral-longitudinal behavior of a vehicle according to this application embodiment are as follows:

[0066] (1) In lane change behavior prediction, for each vehicle, its driving information in the past 3 seconds is input, and a fixed frame rate can be used, such as 10fps, 30 frames, etc. The input information for each frame can include: lateral distance to the lane line, lateral speed, speed difference between the vehicle and the vehicle in front, and longitudinal distance difference.

[0067] Dynamic Bayesian network structures for lane-changing behavior prediction can be like... Figure 6 As shown, its parameter calibration is based on natural driving data in the scenario and trained using the EM algorithm (Expectation-Maximum). The arrows represent the conditional probability relationships between the variables. The latent variables are lane-change area G, lane-change intention C, and lane-change drive ActA, constituting the potential influencing variables of the vehicle's lane-change decision. The manifest variables are lateral distance s to the lane line, lateral speed v, speed difference Dv between the vehicle and the preceding vehicle, and longitudinal distance difference Dl, serving as observable information for the predictor. The predictor outputs the vehicle's lane-change confidence C in the last frame.

[0068] (2) In yielding behavior prediction, for vehicles around a potential conflicting vehicle during the lane change process, input their driving information from the past 2 seconds. A fixed frame rate can be used, such as 10fps or 20fps. The input information for each frame can include: the longitudinal position of the vehicle, the longitudinal speed of the vehicle, the longitudinal position of the conflicting vehicle, and the longitudinal speed of the conflicting vehicle. The time of the last frame should be aligned with the time when the vehicle makes the lane change decision.

[0069] The dynamic Bayesian network structure for yielding behavior prediction can be as follows: Figure 7 As shown, its parameter calibration is based on natural driving data in the scenario and training of the EM algorithm. The arrows represent the conditional probability relationships between the variables. Among them, the latent variables are the speed condition VGap, the distance condition DGap, and the yielding intention Y, which constitute the potential influencing variables of the vehicle's lane-changing decision. The manifest variables are the longitudinal position De of the vehicle, the longitudinal speed Ve of the vehicle, the longitudinal position Do of the conflicting vehicle, the longitudinal speed Vo of the conflicting vehicle, and the yielding and overtaking completion flag M, which serve as the observable information of the predictor. The predictor outputs the yielding confidence Y of the vehicle in the last frame.

[0070] In this embodiment, in the motion model of the vehicle and the motion model of surrounding vehicles, the longitudinal acceleration and the time to complete a sequence for each driving behavior are preset values. The motion model of surrounding vehicles further includes: for longitudinal following behavior, the lead vehicle in the observation area is set to drive at a constant speed, and the following vehicles in the lane adopt an IDM model (Intelligence Driver Model); for longitudinal yielding behavior, a virtual lead vehicle is generated in front of the vehicle, and the longitudinal speed of the vehicle is controlled based on the IDM model. When there is a conflict between surrounding vehicles, the following vehicles yield, and when there is a conflict between surrounding vehicles and the vehicle, the yielding confidence is determined to be greater than a first preset confidence level; for lateral lane changing behavior, when the lane changing vehicle changes to the outermost lane and changes one lane at a time, the lead vehicle in the observation area is set to drive at a constant speed, and the following vehicles in the lane adopt an IDM model. When there is a conflict between the lane changing vehicle and the original lane vehicle, the following vehicles yield, and the lane changing confidence is determined to be a lane changing behavior.

[0071] The first and second pre-set reliability levels can be set according to the actual situation, and no specific restrictions are imposed on them.

[0072] It is understood that, based on the above-described predictions of the lateral and longitudinal behavior of surrounding vehicles, the embodiments of this application can design and generate motion models for the vehicle and surrounding vehicles. Specifically, the design of the motion models for the vehicle and surrounding vehicles in this application is as follows:

[0073] (1) Design of the vehicle motion model:

[0074] The designed autonomous vehicle behavior can include five types: accelerating straight ahead, maintaining a constant speed while straight ahead, decelerating while straight ahead, changing lanes to the left, and changing lanes to the right. The longitudinal acceleration for each behavior is a fixed value, adapted to the scenario; the time to complete each step is also a fixed value, adapted to the scenario.

[0075] (2) Design of the circumferential motion model of the front and rear vehicles:

[0076] The vehicle behavior model can include two types of longitudinal behavior (following and yielding) and two types of lateral behavior (left lane change and right lane change), and the behavior confidence is calculated as shown in the above embodiment. The vehicle motion model parameters are fixed values, the size of which is adapted to the scenario, and the time to complete each step is consistent with the setting of the vehicle motion model.

[0077] For lateral lane-changing behavior, this embodiment of the application can assume that the lane-changing vehicle will change to the outermost lane, changing one lane at a time; during the lane-changing process, the longitudinal speed model is consistent with the longitudinal following. When there is a potential conflict with the vehicle in the original lane, the front and rear order of the two vehicles after the lane change is determined by the start time of the lane change, and the vehicle behind performs longitudinal yielding behavior. When the lane change confidence C is greater than the first preset confidence level (e.g., greater than 80%), it is considered that a lane change has occurred.

[0078] For longitudinal following behavior, this embodiment of the application can assume that the leading vehicle in the lane in the observation area travels at a constant speed, while the following vehicles in the lane adopt an IDM model. For longitudinal yielding behavior, this embodiment of the application can generate a virtual leading vehicle in front of it and control its longitudinal speed based on the IDM model. Therefore, when a conflict occurs between surrounding vehicles, this embodiment of the application can assume that the following vehicle will definitely yield. When a surrounding vehicle conflicts with its own vehicle, it is assumed that the yielding vehicle yields when its yielding confidence Y is greater than a second preset confidence level (e.g., greater than 80%).

[0079] In this embodiment, the behavioral constraints include collision detection, yield verification, drivable area verification, maximum speed verification, and endpoint position verification. The collision detection includes: after the vehicle's driving behavior ends, determining whether the distance and time distance between the vehicle and the vehicles in front and behind in the lane are both greater than constraint values. If they are greater, the driving behavior is deemed feasible; otherwise, it is deemed infeasible. Here, non-vehicle obstacles are virtualized as surrounding vehicles with the same speed as the obstacles. The yield verification includes: when the vehicle performs a lane-changing behavior, if the yield confidence of the conflicting vehicle is less than a preset confidence level, the lane-changing behavior is deemed infeasible; otherwise, it is deemed feasible. Driving area verification includes: if the vehicle's position is not within the driving area after the driving action ends, the driving action is deemed infeasible; otherwise, the driving action is deemed feasible. Maximum speed verification includes: if the vehicle's speed is greater than or equal to the speed limit at the start of acceleration, acceleration is deemed infeasible; otherwise, if the vehicle's speed is greater than or equal to the speed limit at the end of acceleration, the maximum speed of the vehicle is determined to be the speed limit. End point verification includes: if the vehicle's longitudinal position exceeds the target position and is not within the target lane after the driving action ends, the driving action is deemed infeasible.

[0080] Among them, the constraints on the distance and time between the vehicle and the vehicle in front and behind in the lane, as well as the pre-set reliability of yielding and lane changing behaviors, can be specifically set according to the actual situation, and no specific restrictions are imposed on them.

[0081] It is understood that, in the MCTS operation process of this application embodiment, it is necessary to ensure that each generated child node is reachable, that is, based on the set motion model, the vehicle satisfies the feasibility constraints when taking this action. Specifically, the constraints and verification methods for the vehicle's behavior in this application embodiment are as follows:

[0082] (1) Collision test. After the self-vehicle behavior ends, the distance and time distance between it and the vehicles in front and behind in the lane should be greater than the constraint value; otherwise, the behavior is considered infeasible. For other obstacles that are not vehicles, they are virtualized as vehicles with the same speed.

[0083] (2) Yield verification. When performing a lane change, if the yield confidence of a potential conflicting vehicle is less than the preset confidence, such as less than 80%, then the lane change behavior of the vehicle is determined to be infeasible.

[0084] (3) Driving area verification. If the vehicle is not located within the driving area on the map after the action is completed, the action is considered not feasible.

[0085] (4) Maximum speed. If the vehicle speed is greater than or equal to the speed limit at the beginning of the acceleration behavior, the acceleration is considered not feasible; if the vehicle speed is higher than the speed limit at the end of the acceleration behavior, the acceleration is feasible, but the final speed should be equal to the speed limit.

[0086] (5) End point position verification. If the longitudinal position of the vehicle exceeds the target position and is not in the target lane after the action is completed, the action is considered infeasible.

[0087] In this embodiment, a Rollout simulation mechanism is also designed to evaluate the complete strategy for finally reaching the destination. Specifically, during the MCTS iterative operation, if the extended node fails to reach the destination, this embodiment can use Rollout to simulate its remaining behavioral sequence to reach the destination. The Rollout mechanism adopted in this embodiment is a fixed strategy: first, merge into the target lane, and then maintain a constant speed until the destination; wherein, during the simulation, the feasibility constraints in the above embodiments are not considered.

[0088] In this embodiment of the application, an evaluation function is also designed for policy evaluation and node evaluation value updating, and the specific design is as follows:

[0089] (1) Policy Evaluation: For a complete policy, the evaluation function Q is... policy The evaluation function is expressed as a linear weighted sum of four factors: the number of action steps (S), the number of lane changes (H), the distance between the node's lane and the target lane (L), and the node's speed (V). The weighting coefficients are adapted to the scenario, and applying this evaluation function encourages the generation of action sequences that are fast and avoid excessive lane changes. The evaluation function Q is... policy The formula is as follows:

[0090] Q policy = a×S+b×H+c×L+d×V.

[0091] (2) Node Evaluation Update: Based on the evaluation of the policy, update the evaluation values ​​of all nodes in the policy. The UCB1 algorithm (Upper Confidence Bound) is used, and the calculation formula is as follows:

[0092] Q i =α×Q i +β×Q policy ,

[0093]

[0094] Among them, Q i The Q value of node i is updated by combining the previous Q value with the new strategy Q. policy Weighted sum; evaluation function R i Expressed as Qi The weighted sum of the functions of the explored cases of this node; where T i The number of times the strategy passes through node i; the weights are adapted to the scenario.

[0095] In this embodiment of the application, the iteration termination design is also included, including a termination flag: an iteration count limit and a computation time limit; wherein, the iteration termination flag is the flag that takes effect when both the iteration count and computation time are satisfied, and the iteration is terminated, and the flag value is adapted to the scenario.

[0096] In step S103, the driving trajectory of the vehicle is planned based on the first driving behavior of the optimal driving behavior sequence, and after the vehicle executes the first driving behavior based on the driving trajectory, the optimal driving behavior sequence is regenerated until the global planning path is completed.

[0097] Among them, such as Figure 3 As shown, in this embodiment of the application, the final target position and speed of the vehicle behavior in the above steps can be used as the lower-level output and output to the lower-level planner for specific driving trajectory planning.

[0098] It is understood that, in the embodiments of this application, after generating the target optimal driving behavior sequence, the vehicle can only execute the first driving behavior, and after the execution of the behavior is completed, the above decision-making process is repeatedly executed based on the execution result of the vehicle on the driving behavior and the updated scene state information until the vehicle arrives at the target location.

[0099] The following will describe a long-term driving behavior decision-making method applicable to highway and ring road traffic scenarios using a specific algorithm according to an embodiment of this application, such as... Figure 3 As shown, the details are as follows:

[0100] 1. Calculate the target state (first perform a Flener coordinate transformation, and estimate the desired position and motion state of the vehicle after 6 steps based on the map, global path, and the vehicle's and surrounding vehicle's states).

[0101] 2. Prediction of vehicle lateral-longitudinal behavior:

[0102] 2.1 Lane Changing Behavior Prediction: Based on a dynamic Bayesian network, output the confidence score of the lane changing behavior of the weekly vehicle;

[0103] 2.2 Yielding Behavior Prediction: Based on a dynamic Bayesian network, output the confidence score of "yielding" behavior when a vehicle is going straight.

[0104] 3. Long-term driving behavior decision-making based on MCTS:

[0105] 3.1 Design of vehicle motion model (accelerating straight, constant speed straight, decelerating straight, changing lanes to the left, changing lanes to the right);

[0106] 3.2 Design of the vehicle motion model (longitudinal: following, yielding; lateral: left / right lane change), wherein the behavior prediction is given by the aforementioned "vehicle lateral-longitudinal behavior prediction" module;

[0107] 3.3 Node State Update: A linked list is established for each lane to store the motion state of each vehicle. When the state is updated, it is updated based on the aforementioned vehicle motion model and surrounding vehicle motion model.

[0108] 3.4. Behavioral feasibility verification (collision test, drivable area test, etc.);

[0109] 3.5 Rollout simulation mechanism design (fixed strategy: first merge into the target lane, then drive at a constant speed in a straight line to the target);

[0110] 3.6 Evaluation function design (strategy evaluation and node evaluation update);

[0111] 3.7 Iterative Termination Design (Termination Criteria: Iteration Limit, Computation Time Limit).

[0112] 4. Output: Calculate the final state of the first step of the optimal sequence and transform it back to Cartesian coordinates for output.

[0113] 5. Repeated Execution: After the vehicle driving behavior is completed, repeat steps 1 to 4. If the behavior is found to be infeasible during the execution process, adopt a risk-avoidance driving strategy.

[0114] In summary, this application embodiment can acquire scene state information, including the motion state of the vehicle and surrounding vehicles, the road geometry, traffic regulations, and the planned destination. Then, it optimizes and solves the long-term driving behavior sequence based on the MCTS algorithm. During the iterative solution of driving behavior, the lane-changing and yielding intentions of surrounding vehicles in the lateral and longitudinal directions are identified by corresponding dynamic Bayesian networks, and collision constraints, road structure, and regulatory constraints are considered to ensure the feasibility of the generated driving strategy. After generating the target behavior sequence, the vehicle only executes the first behavior. After this behavior is completed, the above decision-making process is repeatedly executed based on the updated scene state information until the vehicle reaches the destination. This application embodiment is mainly applicable to intelligent driving vehicle applications in structured road scenarios such as highways and urban ring roads, and has the function of realizing multi-step decision-making and generating highly feasible optimal solutions, which helps to comprehensively improve the safety and efficiency of intelligent vehicle operation.

[0115] The long-term driving behavior decision-making method proposed in the embodiments of this application, applicable to highway and ring road traffic scenarios, can achieve multi-step decision-making for driving behaviors such as lane change gap selection and lane overtaking by coupling the optimal decision objective and surrounding vehicle behavior response prediction in multi-lane driving scenarios such as highways and ring roads, based on MCTS, and considering the optimal decision objective and surrounding vehicle behavior response prediction in highway scenarios. It has a forward-looking nature. At the same time, based on feasibility judgment, it provides the optimal solution for long-term driving behavior sequence planning, taking into account both safety and efficiency, and meeting the needs of general driving.

[0116] Next, referring to the accompanying drawings, a long-term driving behavior decision-making device suitable for highway and ring road traffic scenarios is described according to an embodiment of this application.

[0117] Figure 8 This is a block diagram of a long-term driving behavior decision-making device applicable to highway and ring road traffic scenarios, according to an embodiment of this application.

[0118] like Figure 8 As shown, the long-term driving behavior decision-making device 10, applicable to highway and ring road traffic scenarios, includes: an information acquisition module 100, a decision-making module 200, and a control module 300.

[0119] The information acquisition module 100 is used to acquire the global planned path of the vehicle, the current motion state of the vehicle, and the current motion state of all surrounding vehicles in the area; the decision module 200 is used to generate the optimal driving behavior sequence of the vehicle based on the global planned path, the current motion state of the vehicle, and the current motion state of all surrounding vehicles in the area; the control module 300 is used to plan the driving trajectory of the vehicle based on the first driving behavior of the optimal driving behavior sequence, and after controlling the vehicle to execute the first driving behavior based on the driving trajectory, it regenerates the optimal driving behavior sequence until the global planned path is completed.

[0120] In this embodiment, the decision module 200 is used to: take the current motion state of the vehicle and the current motion state of all surrounding vehicles in the area as the root node, sequentially select the optimal child node until the leaf node; at the leaf node, calculate the new child node state of the vehicle under all driving behaviors, and perform reachability calculation on each new child node state based on the motion model of the vehicle and the motion model of all surrounding vehicles in the area to obtain reachable nodes; among all reachable child nodes, randomly select a child node as an extension node, and start from the extension node to reach the target state based on the Rollout strategy to obtain simulation results; perform backpropagation based on the simulation results, update the evaluation values ​​of all nodes on the child node path until the iteration stopping condition is met, and determine the optimal path based on the evaluation values ​​of all nodes, and generate the optimal driving behavior sequence based on the driving behaviors corresponding to the nodes on the optimal path.

[0121] In this embodiment of the application, each node stores the motion state of the vehicle and all surrounding vehicles in the area, and updates the motion state of the vehicle and all surrounding vehicles in the area in each node based on the motion model of the vehicle and the motion model of all surrounding vehicles in the area.

[0122] In this embodiment, the decision module 200 is further configured to: predict the lateral and longitudinal behaviors of all surrounding vehicles in the region based on the current motion state of the vehicle and the current motion state of all surrounding vehicles in the region, and obtain lateral and longitudinal behavior prediction results; match the motion models of all surrounding vehicles in the region based on the lateral and longitudinal behavior prediction results, and verify whether the state of each new child node satisfies the behavior constraints of the vehicle based on the motion model of the vehicle and the motion models of all surrounding vehicles in the region; and determine the reachability of the node when the behavior constraints of the vehicle are satisfied.

[0123] In this embodiment, the decision module 200 is further configured to: input the current motion state and the predicted current motion state of all surrounding vehicles in the area into a pre-trained dynamic Bayesian network for lane-changing behavior prediction, and output the confidence score of lane-changing behavior of all surrounding vehicles in the area. The variable information in the dynamic Bayesian network for lane-changing behavior prediction includes lane-changing area, lane-changing intention, lane-changing drive, lateral distance of lane line, lateral speed, speed difference with the preceding vehicle, and longitudinal distance difference; input the current motion state and the predicted current motion state of all surrounding vehicles in the area into a pre-trained dynamic Bayesian network for yielding behavior prediction, and output the yielding confidence score. The variable information in the dynamic Bayesian network for yielding behavior prediction includes speed condition, distance condition, yielding intention, longitudinal position of the vehicle, longitudinal speed of the vehicle, longitudinal position of the following vehicle, longitudinal speed of the following vehicle, and yielding / cutting completion flags.

[0124] In this embodiment, the decision module 200 is further configured to: for longitudinal following behavior, set the leading vehicle in the observation area to drive at a constant speed, and the following vehicles in the lane to adopt the IDM model; for longitudinal yielding behavior, generate a virtual leading vehicle in front of the vehicle, and control the longitudinal speed of the vehicle based on the IDM model, and when there is a conflict between surrounding vehicles, the following vehicles yield, and when there is a conflict between surrounding vehicles and the vehicle itself, determine that the surrounding vehicles yield when the yielding confidence is greater than a first preset confidence level; for lateral lane changing behavior, when the lane changing vehicle changes to the outermost lane and changes one lane at a time, set the leading vehicle in the observation area to drive at a constant speed, and the following vehicles in the lane to adopt the IDM model, and when there is a conflict between the lane changing vehicle and the original lane vehicle, the following vehicles yield, and determine that the lane changing vehicle has lane changing behavior when the lane changing confidence is greater than a second preset confidence level.

[0125] In this embodiment, the behavioral constraints include collision detection, yield verification, drivable area verification, maximum speed verification, and endpoint position verification. The collision detection includes: after the vehicle's driving behavior ends, determining whether the distance and time distance between the vehicle and the vehicles in front and behind in the lane are both greater than constraint values. If they are greater, the driving behavior is deemed feasible; otherwise, it is deemed infeasible. Here, non-vehicle obstacles are virtualized as surrounding vehicles with the same speed as the obstacles. The yield verification includes: when the vehicle performs a lane-changing behavior, if the yield confidence of the conflicting vehicle is less than a preset confidence level, the lane-changing behavior is deemed infeasible; otherwise, it is deemed feasible. Driving area verification includes: if the vehicle's position is not within the driving area after the driving action ends, the driving action is deemed infeasible; otherwise, the driving action is deemed feasible. Maximum speed verification includes: if the vehicle's speed is greater than or equal to the speed limit at the start of acceleration, acceleration is deemed infeasible; otherwise, if the vehicle's speed is greater than or equal to the speed limit at the end of acceleration, the maximum speed of the vehicle is determined to be the speed limit. End point verification includes: if the vehicle's longitudinal position exceeds the target position and is not within the target lane after the driving action ends, the driving action is deemed infeasible.

[0126] In this embodiment of the application, the driving behavior includes any one of accelerating straight, driving straight at a constant speed, decelerating straight, changing lanes to the left, and changing lanes to the right.

[0127] It should be noted that the foregoing explanation of the long-term driving behavior decision-making method embodiment applicable to highway and ring road traffic scenarios also applies to the long-term driving behavior decision-making device of this embodiment applicable to highway and ring road traffic scenarios, and will not be repeated here.

[0128] The long-term driving behavior decision-making device proposed in the embodiments of this application, applicable to highway and ring road traffic scenarios, can, based on MCTS, couple the optimal decision-making objective in highway scenarios with the prediction of surrounding vehicle behavior responses to achieve multi-step decision-making for driving behaviors such as lane change interval selection and lane overtaking, and has a forward-looking nature. At the same time, based on feasibility judgment, it provides the optimal solution for long-term driving behavior sequence planning, taking into account both safety and efficiency, and meeting the needs of general driving.

[0129] Figure 9 A schematic diagram of the structure of a vehicle provided in an embodiment of this application. The vehicle may include:

[0130] The memory 901, the processor 902, and the computer program stored on the memory 901 and capable of running on the processor 902.

[0131] When the processor 902 executes the program, it implements the long-term driving behavior decision-making method for highway and ring road traffic scenarios provided in the above embodiments.

[0132] Furthermore, the vehicle also includes:

[0133] Communication interface 903 is used for communication between memory 901 and processor 902.

[0134] The memory 901 is used to store computer programs that can run on the processor 902.

[0135] The memory 901 may include high-speed RAM (Random Access Memory) memory, and may also include non-volatile memory, such as at least one disk storage.

[0136] If the memory 901, processor 902, and communication interface 903 are implemented independently, then the communication interface 903, memory 901, and processor 902 can be interconnected via a bus to complete communication between them. The bus can be an ISA (Industry Standard Architecture) bus, a PCI (Peripheral Component Interconnect) bus, or an EISA (Extended Industry Standard Architecture) bus, etc. The bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 9 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0137] Optionally, in a specific implementation, if the memory 901, processor 902, and communication interface 903 are integrated on a single chip, then the memory 901, processor 902, and communication interface 903 can communicate with each other through an internal interface.

[0138] The processor 902 may be a CPU (Central Processing Unit), an ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of this application.

[0139] This application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described long-term driving behavior decision-making method applicable to highway and ring road traffic scenarios.

[0140] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0141] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0142] Any process or method described in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more N executable instructions for implementing custom logic functions or processes, and the scope of the preferred embodiments of this application includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as should be understood by those skilled in the art to which embodiments of this application pertain.

[0143] It should be understood that the various parts of this application can be implemented using hardware, software, firmware, or a combination thereof. In the above embodiments, the N steps or methods can be implemented using software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (FPGAs), field-programmable gate arrays (FPGAs), etc.

[0144] Those skilled in the art will understand that all or part of the steps of the methods in the above embodiments can be implemented by a program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, the program includes one or a combination of the steps of the method embodiments.

[0145] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A long-term driving behavior decision-making method applicable to highway and ring road traffic scenarios, characterized in that, Includes the following steps: Obtain the vehicle's global planned path, the vehicle's current motion state, and the current motion state of all surrounding vehicles within the area; The optimal driving behavior sequence of the vehicle is generated based on the global planning path, the current motion state of the vehicle, and the current motion states of all surrounding vehicles in the area. This includes: taking the current motion state of the vehicle and the current motion states of all surrounding vehicles in the area as the root node, and sequentially selecting the optimal child node until the leaf node. At the leaf node, the new child node states of the vehicle under all driving behaviors are calculated, and the reachability of each new child node state is calculated based on the vehicle's motion model and the motion models of all surrounding vehicles in the area to obtain reachable nodes, including: Based on the current motion state of the vehicle and the current motion states of all surrounding vehicles within the area, the lateral and longitudinal behaviors of all surrounding vehicles within the area are predicted to obtain lateral and longitudinal behavior prediction results, including: The current motion state and the predicted current motion states of all surrounding vehicles in the area are input into a pre-trained dynamic Bayesian network for lane change behavior prediction, and the network outputs the confidence scores of lane change behavior of all surrounding vehicles in the area. The variable information in the dynamic Bayesian network for lane change behavior prediction includes lane change area, lane change intention, lane change drive, lane line lateral distance, lateral speed, speed difference with the vehicle in front, and longitudinal distance difference. The current motion state and the predicted current motion states of all surrounding vehicles in the area are input into a pre-trained yielding behavior prediction dynamic Bayesian network, and the yielding confidence is output. The variable information in the yielding behavior prediction dynamic Bayesian network includes speed conditions, distance conditions, yielding intention, longitudinal position of the vehicle, longitudinal speed of the vehicle, longitudinal position of the following vehicle, longitudinal speed of the following vehicle, and yielding or overtaking completion indicators. Based on the lateral and longitudinal behavior prediction results, match the motion models of all surrounding vehicles in the region, and verify whether the state of each new child node satisfies the behavior constraints of the vehicle based on the motion model of the vehicle and the motion models of all surrounding vehicles in the region. When the behavioral constraints of the vehicle are met, the node is determined to be reachable; The behavioral constraints include yield verification, which includes: when the vehicle performs a lane-changing behavior, if the yield confidence of the vehicle conflicting with the vehicle is less than a preset confidence, the lane-changing behavior is determined to be infeasible; otherwise, the lane-changing behavior is determined to be feasible. as well as The vehicle's driving trajectory is planned based on the first driving action of the optimal driving behavior sequence, and after the vehicle executes the first driving action based on the driving trajectory, the optimal driving behavior sequence is regenerated until the global planning path is completed.

2. The method according to claim 1, characterized in that, The step of generating the optimal driving behavior sequence for the vehicle based on the global planning path, the current motion state of the vehicle, and the current motion states of all surrounding vehicles within the area further includes: Among all reachable child nodes, a child node is randomly selected as the extension node, and the target state is reached from the extension node based on the Rollout strategy to obtain the simulation results; Backpropagation is performed based on the simulation results to update the evaluation values ​​of all nodes on the child node path until the iteration stopping condition is met. The optimal path is determined based on the evaluation values ​​of all nodes, and the optimal driving behavior sequence is generated based on the driving behavior corresponding to the nodes on the optimal path.

3. The method according to claim 2, characterized in that, in, Each node stores the motion state of the vehicle and all surrounding vehicles within the area, and updates the motion state of the vehicle and all surrounding vehicles within the area in each node based on the motion model of the vehicle and the motion model of all surrounding vehicles within the area.

4. The method according to claim 1, characterized in that, in, In the motion model of the vehicle and the motion model of the surrounding vehicles, the longitudinal acceleration and the time to complete a sequence for each driving action are preset values. The motion model of the surrounding vehicles also includes: For longitudinal following behavior, the leading vehicle in the lane in the observation area is driven at a constant speed, while the following vehicles in the lane are driven using the IDM model. For longitudinal yielding behavior, a virtual vehicle in front is generated in front of the vehicle, and the longitudinal speed of the vehicle is controlled based on the IDM model. When there is a conflict between surrounding vehicles, the rear vehicle yields. When there is a conflict between the surrounding vehicle and the vehicle, the surrounding vehicle is determined to yield when the yielding confidence is greater than the first preset confidence. For lateral lane change behavior, when a lane-changing vehicle changes to the outermost lane and changes one lane at a time, the leading vehicle in the lane in the observation area travels at a constant speed, and the following vehicles in the lane adopt the IDM model. When there is a conflict between the lane-changing vehicle and the original lane vehicle, the following vehicle yields, and when the lane change confidence is greater than the second preset confidence, it is determined that the lane-changing vehicle has engaged in lane change behavior.

5. The method according to claim 4, characterized in that, The behavioral constraints also include collision detection, drivable area verification, maximum speed verification, and endpoint location verification, among which... The collision test includes: after the driving behavior of the vehicle ends, determining whether the distance and time distance between the vehicle and the vehicle in front and behind in the lane are both greater than the constraint value. If they are greater, the driving behavior is determined to be feasible; otherwise, the driving behavior is determined to be infeasible. In this case, obstacles that are not vehicles are virtualized as surrounding vehicles with the same speed as the obstacles. The driving area verification includes: after the driving behavior of the vehicle ends, if the vehicle is not in the driving area, the driving behavior is determined to be infeasible; otherwise, the driving behavior is determined to be feasible. The maximum speed verification includes: when the vehicle's acceleration begins, if the vehicle's speed is greater than or equal to the speed limit at the current position, then acceleration is determined to be infeasible; otherwise, after the acceleration ends, if the vehicle's speed is greater than or equal to the speed limit at the current position, then the maximum speed of the vehicle is determined to be the speed limit. The endpoint position verification includes: if the longitudinal position of the vehicle exceeds the target position and is not within the target lane after the driving behavior of the vehicle ends, the driving behavior is determined to be infeasible.

6. The method according to any one of claims 1-5, characterized in that, The driving behavior includes any one of the following: accelerating straight, driving straight at a constant speed, decelerating straight, changing lanes to the left, or changing lanes to the right.

7. A long-term driving behavior decision-making device suitable for highway and ring road traffic scenarios, characterized in that, include: The information acquisition module is used to acquire the global planned path of the vehicle, the current motion state of the vehicle, and the current motion state of all surrounding vehicles in the area; The decision module is used to generate the optimal driving behavior sequence of the vehicle based on the global planning path, the current motion state of the vehicle and the current motion state of all surrounding vehicles in the area, including: taking the current motion state of the vehicle and the current motion state of all surrounding vehicles in the area as the root node, and sequentially selecting the optimal child node until the leaf node. At the leaf node, the new child node states of the vehicle under all driving behaviors are calculated, and the reachability of each new child node state is calculated based on the vehicle's motion model and the motion models of all surrounding vehicles in the area to obtain reachable nodes, including: Based on the current motion state of the vehicle and the current motion states of all surrounding vehicles within the area, the lateral and longitudinal behaviors of all surrounding vehicles within the area are predicted to obtain lateral and longitudinal behavior prediction results, including: The current motion state and the predicted current motion states of all surrounding vehicles in the area are input into a pre-trained dynamic Bayesian network for lane change behavior prediction, and the network outputs the confidence scores of lane change behavior of all surrounding vehicles in the area. The variable information in the dynamic Bayesian network for lane change behavior prediction includes lane change area, lane change intention, lane change drive, lane line lateral distance, lateral speed, speed difference with the vehicle in front, and longitudinal distance difference. The current motion state and the predicted current motion states of all surrounding vehicles in the area are input into a pre-trained yielding behavior prediction dynamic Bayesian network, and the yielding confidence is output. The variable information in the yielding behavior prediction dynamic Bayesian network includes speed conditions, distance conditions, yielding intention, longitudinal position of the vehicle, longitudinal speed of the vehicle, longitudinal position of the following vehicle, longitudinal speed of the following vehicle, and yielding or overtaking completion indicators. Based on the lateral and longitudinal behavior prediction results, match the motion models of all surrounding vehicles in the region, and verify whether the state of each new child node satisfies the behavior constraints of the vehicle based on the motion model of the vehicle and the motion models of all surrounding vehicles in the region. When the behavioral constraints of the vehicle are met, the node is determined to be reachable; The behavioral constraints include yield verification, which includes: when the vehicle performs a lane-changing behavior, if the yield confidence of the vehicle conflicting with the vehicle is less than a preset confidence, the lane-changing behavior is determined to be infeasible; otherwise, the lane-changing behavior is determined to be feasible. as well as The control module is used to plan the driving trajectory of the vehicle based on the first driving behavior of the optimal driving behavior sequence, and after controlling the vehicle to execute the first driving behavior based on the driving trajectory, regenerate the optimal driving behavior sequence until the global planning path is completed.

8. A vehicle, characterized in that, include: The memory, the processor, and the computer program stored in the memory and executable on the processor, the processor executing the program to implement the long-term driving behavior decision-making method for highway and ring road traffic scenarios as described in any one of claims 1-5.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, The program is executed by the processor to implement the long-term driving behavior decision-making method applicable to highway and ring road traffic scenarios as described in any one of claims 1-5.

Images