An automatic driving vehicle hierarchical game decision method for low-speed merging scene

Abstract

The application provides a kind of automatic driving vehicle hierarchical game decision method for low-speed merging scene, first, the merging process of automatic driving vehicle is divided into multiple stages by using finite state machine;Second, through real-time gap detection, filter the feasible safe merging gap, and determine the target vehicle participating in interaction;Then, the interaction behavior between automatic driving vehicle and target vehicle is modeled as a game problem, and based on the current state information of both sides, the feasible motion trajectory is planned in the action space;Next, according to the designed cost function, the cost corresponding to the trajectory combination of both is calculated, forming a double cost matrix, finding Nash equilibrium in the matrix, and outputting it as the current optimal strategy;Finally, the trajectory under the optimal strategy is collision predicted, the collision risk of automatic driving vehicle is detected in real time, the emergency brake is triggered when necessary to avoid the target vehicle, and the cycle is re-entered to select a new safe merging gap for a new round of real-time gap detection. The scheme of the application significantly improves the robustness, safety and operation efficiency of the automatic driving system, effectively improves the overall traffic efficiency and riding experience.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of automatic driving vehicle decision control, and particularly relates to a hierarchical game decision method for automatic driving vehicles in a low-speed merging scene. BACKGROUND

[0002] With the continuous development of automatic driving technology, the basic cruise control of vehicles on structured roads (such as adaptive cruise, lane keeping, etc.) has been relatively mature. However, in complex scenes that require high-intensity interaction with other traffic participants, especially in low-speed merging scenes, the decision-making ability of automatic driving vehicles still faces great challenges. In a low-speed merging scene, the merging vehicle needs to choose a suitable gap and may force the vehicles on the target lane to slow down in order to safely merge into the traffic flow, and interaction and / or cooperation often need to be carried out between the merging vehicle and the vehicles driving on the target lane.

[0003] Currently, some patents have explored the decision control of automatic driving vehicles in merging scenes. CN120088978A discloses a vehicle-road cooperative ramp merging control method based on reinforcement learning, which realizes the system merging control of intelligent networked vehicles equipped with intelligent agents in the ramp and main line merging area. However, intelligent networked vehicles have not yet been fully popularized in the current real traffic environment, and it is difficult to achieve coordinated control of interactive vehicles in the cloud. CN119953404A discloses an intelligent vehicle ramp merging control method based on driving style recognition, which effectively avoids the occurrence of vehicle collisions in the ramp merging area. However, due to the dependence of the clustering pre-training model on real traffic data, the cost of landing automatic driving vehicles is increased. CN120412313A discloses a freeway diverging area forced lane change decision method for intelligent networked environments, which realizes safe, efficient, and comfortable lane change decision-making of automatic driving vehicles in the forced lane change scene of the diverging area. CN119889077A discloses a highway vehicle ramp merging decision method based on insertable gaps, which effectively solves the lane change problem of automatic driving vehicles when merging at the ramp of the highway. However, the above two algorithms plan trajectories through self-optimization and do not fully consider the bidirectional interaction between vehicles in the low-speed merging scene, i.e., only consider the influence of other vehicles on the vehicle itself, without considering the influence of the vehicle itself on other vehicles.

[0004] Game theory is a branch of mathematics that can effectively describe the interaction between agents, so some researchers have applied game theory methods to model vehicle behavior interaction. The core of game theory is to provide a mathematical model of strategic decision-making and formalize the gains and losses faced by participants, which makes game participants decide whether to compete or cooperate with others. Patent CN120482094A discloses an automatic driving vehicle longitudinal speed control system and method based on game theory, which optimizes the game between the ego vehicle and other manually driven vehicles, and improves the ability of the automatic driving vehicle to respond to the complex behavior of other manually driven vehicles. However, this method is only applicable to longitudinal speed control in simple scenarios. For strong interaction scenarios, only considering longitudinal control results in low strategy accuracy and efficiency, making it difficult to apply to low-speed merging scenarios.

[0005] Therefore, the existing method still has deficiencies in handling the interaction decision of an automatic driving vehicle in a low-speed merging scenario, and cannot obtain higher traffic efficiency at a low cost while ensuring safety. SUMMARY

[0006] To solve the above problems, the present application provides an automatic driving vehicle hierarchical game decision method for a low-speed merging scenario, which can realize the automatic driving vehicle merging decision function considering decision intelligence, planning feasibility and system safety.

[0007] In one aspect, the present application provides an automatic driving vehicle hierarchical game decision method for a low-speed merging scenario, which comprises:

[0008] Step S1, the automatic driving vehicle receives a merging instruction and prepares to enter the merging phase;

[0009] Step S2, vehicle gap determination, the automatic driving vehicle evaluates the state of surrounding vehicles, selects the optimal safe vehicle gap, and determines the target vehicle for game theory trajectory planning;

[0010] Step S3, interactive game between the automatic driving vehicle and the target vehicle to obtain a safe and feasible Nash equilibrium solution;

[0011] Step S4, the automatic driving vehicle executes the control instruction in the Nash equilibrium solution and performs real-time dynamic detection on the collision risk during the lane changing process;

[0012] Step S5, when the game interaction fails, the automatic driving vehicle returns to the vehicle gap determination stage to find a new merging opportunity while maintaining a safe following distance;

[0013] Step S6, the automatic driving vehicle safely merges into the target vehicle and completes the merging task.

[0014] Further, the step S2 specifically comprises:

[0015] Selecting optimal inter-vehicle gap by minimizing risk indicator , which is calculated as:

[0016] ;

[0017] wherein, denotes the longitudinal position coordinate of the ego vehicle, denotes the longitudinal position coordinate of the front vehicle at the gth inter-vehicle gap, denotes the longitudinal position coordinate of the rear vehicle at the gth inter-vehicle gap, , denotes the minimum safety gap.

[0018] Further, the interactive game in the step S3 comprises:

[0019] First, according to the current state information of the ego vehicle, i.e. the merging vehicle, and the target vehicle, i.e. the rear vehicle behind the optimal inter-vehicle gap, the feasible motion trajectories of the two vehicles in the action space are planned respectively; second, according to the designed cost function, the cost of the combination of different trajectories of the two vehicles is calculated respectively, and a double cost matrix is obtained; third, the Nash equilibrium is found in the established cost matrix; then, the obtained optimal strategy is checked for safety protection; finally, if there is a collision in the predicted trajectory of the vehicle, the ego vehicle selects emergency braking to terminate the lane change to avoid the straight-ahead target vehicle, otherwise, the obtained Nash equilibrium is output as the current optimal strategy.

[0020] Further, the planning of the feasible motion trajectory in the action space comprises:

[0021] planning a set of sampled vehicle motion trajectories as the action space of the vehicle within a certain time , the longitudinal and lateral motion of the two vehicles is modeled based on a 5th order polynomial:

[0022] ;

[0023] wherein, denotes the vehicle trajectory, i.e. the position coordinates of the vehicle in the longitudinal and lateral directions changing with time, denotes the coefficients of the polynomial equation, , is determined by the initial state T init = [ x init , y init , x ˙ init , y ˙ init , x ¨ init , y ¨ init ] and the terminal state T term = [ x term , y term , x ˙ term , y ˙ term , x ¨ term , y ¨ term ] of the vehicle trajectory, wherein, denote the position of the vehicle in the longitudinal and lateral directions, respectively, denote the speed of the vehicle in the longitudinal and lateral directions, respectively, These represent the vehicle's acceleration in the longitudinal and lateral directions, respectively.

[0024] Discretize the vehicle position at the termination time within the feasible region, and the planning resolution for merging the vehicle AV and the following vehicle SV is: ,in Let represent the intervals of the discrete positions of vehicle j in the longitudinal and lateral directions at the termination time, respectively. Then, the longitudinal velocity of the vehicle at the termination time is discretized as... ,in This represents the number of discrete longitudinal velocities, with a discrete interval of 1, yielding feasible motion trajectories for the merging vehicle and the following vehicle within the action space. Its discrete quantity is: ,in, and These represent the number of discretized vehicle positions at the termination time in the longitudinal and lateral directions, respectively. Based on scene and kinematic constraints, the planned motion trajectories in the action space are filtered out, that is, all motion trajectories that satisfy the upper and lower limits of vehicle state variables in the scene are selected as feasible action spaces. The upper and lower limits of vehicle state variables include the vehicle's lateral and longitudinal coordinates, maximum longitudinal and lateral velocities, and maximum longitudinal and lateral accelerations.

[0025] Furthermore, the inbound vehicle AV and the target vehicle SV are identified as the participants in the game, and the vehicle running cost for each vehicle in the established game model at time step k is designed:

[0026] ;

[0027] in, Let represent the travel costs of the merging vehicle and the following vehicle from time k to time k+1, respectively. These represent the states of the merging vehicle, the preceding vehicle FV, and the following vehicle SV at time k, respectively. The states include their longitudinal and lateral positions and velocities. s k j = [ x k j , y k j , x ˙ k j , y ˙ k j ] T , , These represent the control inputs for the selected trajectories of the two vehicles in the motion space, namely the longitudinal and lateral accelerations. u k j = [ x ¨ k j , y ¨ k j ] T , , r k AV = [ r 1 AV , r 2 AV , r 3 AV , r 4 AV ] and r k SV = [ r 1 SV , r 2 SV , r 3 SV , r 4 SV , r 5 SV ] These represent the combinations of the various indices of the cost functions for the two vehicles. ω AV = [ w 1 AV , w 2 AV , w 3 AV , w 4 AV ] T and ω SV = [ w 1 SV , w 2 SV , w 3 SV , w 4 SV , w 5 SV ] T These represent the cost function weight vectors for the two vehicles, which are flexibly adjusted according to the driving style preferences of the vehicles.

[0028] The cost function indicators for the merging vehicle are defined as:

[0029] represents the collision safety component of the merging vehicle, i.e. if any collision occurs between the merging vehicle, the preceding vehicle and the following vehicle, then , otherwise The weight of the collision safety component is greater than the weights of the efficiency and comfort cost components;

[0030] represents the safety constraint component of the merging vehicle, i.e. the merging vehicle should maintain a safe distance from the end of the lane in which it is located:

[0031] ;

[0032] wherein represents the longitudinal distance of the merging vehicle from the end of the lane in which it is located, represents the lateral position coordinate of the merging vehicle, represents the minimum longitudinal safe distance of the merging vehicle from the end of the lane in which it is located;

[0033] represents the efficiency component of the merging vehicle, which is defined as:

[0034] ;

[0035] wherein represents the lane width of the road;

[0036] represents the ride comfort component of the merging vehicle, which is defined as:

[0037] ;

[0038] wherein and represent the longitudinal and lateral acceleration values of the merging vehicle, respectively, and represent the maximum longitudinal and lateral acceleration values of the merging vehicle, respectively;

[0039] Similarly, the cost function indicators for the following vehicle are defined as:

[0040] represents the collision safety component of the following vehicle, i.e. if any collision occurs between the merging vehicle, the preceding vehicle and the following vehicle, then , otherwise The weight of the collision safety component is greater than the weights of the efficiency and comfort cost components;

[0041] represents the safety constraint component of the rear vehicle, i.e., the rear vehicle should maintain a safe distance from the front vehicle:

[0042] ;

[0043] wherein, represents the longitudinal distance of the rear vehicle from the front vehicle, and respectively represent the lateral position coordinate values of the front vehicle and the rear vehicle,

[0044] represents the efficiency component of the rear vehicle, which is defined as:

[0045] ;

[0046] wherein, represents the longitudinal distance of the rear vehicle from the end of the merging area;

[0047] represents the ride comfort component of the rear vehicle, which is defined as:

[0048] ;

[0049] wherein, and respectively represent the acceleration values of the rear vehicle in the longitudinal and lateral directions, and respectively represent the maximum acceleration values of the rear vehicle in the longitudinal and lateral directions;

[0050] represents the penalty term of the rear vehicle deviating from the lane centerline, which is defined as: wherein, represents the lateral position coordinate of the rear vehicle;

[0051] For the trajectory planned in the action space, the total cost of completing the trajectory tracking task is obtained by accumulating the running costs of multiple steps:

[0052] ;

[0053] wherein, and respectively represent the discrete time steps required for the merging vehicle and the rear vehicle to complete the task according to the trajectory;

[0054] Finally, a double cost matrix of is obtained according to the total cost.

[0055] Further, a Nash equilibrium is sought in the established cost matrix, including:

[0056] First, the iterative elimination of strictly dominated strategies is used to find a pure strategy equilibrium, if not found, then the Lemke-Howson method is used to find a mixed strategy in the game, define a probability distribution combination :

[0057] ;

[0058] wherein, , denotes the probability distribution of the vehicle j selecting the motion trajectory , p i j ∈ [ 0 , 1 ] , i ∈ { 1 ,..., M j } , j ∈ { AV , SV } denotes the probability of the vehicle j selecting the motion trajectory , the probability distribution is obtained by the Lemke-Howson method.

[0059] Further, the step S4 of dynamically detecting the collision risk in the lane changing process in real time comprises:

[0060] Based on two indexes, the collision time and the modified headway time are corrected, and the comprehensive lane-cutting risk index CRI in the lane changing process is designed by using the softmax function:

[0061] ;

[0062] wherein, and respectively denote the Sigmoid regularization index of the modified collision time MTTC and the modified headway time MTHW:

[0063] ;

[0064] wherein, are both sensitivity parameters used to adjust the collision risk, MTTC denotes the modified collision time, and MTHW denotes the modified headway time:

[0065] ;

[0066] ;

[0067] wherein, and respectively denote the distance from the autonomous vehicle to the conflict area and the distance from the rear vehicle to the conflict area, and respectively denote the speed of the autonomous vehicle and the speed of the rear vehicle, denotes that the autonomous vehicle has entered the target lane.

[0068] In another aspect, the present application also provides an electronic device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the steps of the method as described above when executing the computer program.

[0069] In another aspect, the present application also provides a computer readable storage medium comprising a stored computer program, wherein the computer program, when executed by a processor, controls the device in which the storage medium is located to perform the steps of the method as described above.

[0070] In summary, due to the adoption of the above technical solutions, the present application has the following beneficial technical effects:

[0071] The present application proposes a low-speed merging scenario automatic driving hierarchical game decision-making method considering collision risk, which divides the automatic driving vehicle merging maneuver into six discrete stages through the merging cut-in decision-making framework of the upper layer finite state machine, and designs rule-based conversion logic, so that the merging vehicle can continue to enter the next interactive state cycle after the interaction fails, which effectively overcomes the limitations of poor adaptability and rigid decision-making of traditional rule-based or single optimization methods in low-speed dynamic strong interaction environment. In the trajectory planning stage of the lower layer game theory, the interaction between the automatic driving vehicle and the target vehicle is modeled as a Nash equilibrium solving problem, and a feasible action space is designed for the two vehicles, and a multi-objective cost function considering safety, efficiency and comfort is designed, and further through trajectory prediction and collision detection, redundant protection is realized to ensure that even in the case of game interaction failure (no safe and feasible optimal strategy output), emergency braking is entered to avoid collision, and then the automatic driving vehicle game returns to the distance selection stage. Compared with the prior art, the method significantly improves the rationality and robustness of the decision-making process while ensuring real-time, reduces the accident risk in the merging process, and significantly improves the decision-making intelligence and system safety of the automatic driving vehicle in the low-speed merging scenario;

[0072] The present application realizes the unity of computational efficiency and practicality through hierarchical structure and adaptive mechanism, proposes a closed-loop decision-making mechanism based on finite state machine and recursive game reinitialization, which can dynamically identify the optimal merging gap and only interact with key vehicles (such as "rear vehicle"), greatly reducing the computational complexity in multi-vehicle scenarios; at the same time, through the discretization design of the action space and the polynomial trajectory planning method, the computational burden is effectively controlled under the premise of ensuring the smoothness and dynamics feasibility of the trajectory, overcoming the low search efficiency and real-time deficiency of existing methods in high-dimensional continuous space. The mechanism makes the automatic driving system have strong engineering realizability and expansibility, providing a technical foundation for the actual deployment of the automatic driving system;

[0073] The application introduces adjustable weight parameters in the design of the cost function, such as cost function weight and collision index threshold, which can flexibly adjust the decision preference according to different driver types (such as aggressive type and conservative type) or scene requirements, and enhance the personalized adaptability of the system; in addition, by optimizing the efficiency component and the comfort component, the merging time and acceleration mutation are effectively reduced under the premise of ensuring safety, the road traffic efficiency and the riding comfort are improved, which helps to relieve traffic congestion and reduce energy consumption, compared with the traditional decision method which takes single safety as the primary target, the application has significant advantages in humanized design and overall traffic efficiency, has good driving adaptability and human-machine cooperation potential, and improves traffic efficiency and riding experience. BRIEF DESCRIPTION OF DRAWINGS

[0074] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and those skilled in the art can obtain other drawings according to these drawings without creative labor.

[0075] Figure 1 is a flowchart of an automatic driving vehicle hierarchical game decision method for low-speed merging scene provided by an embodiment of the present application;

[0076] Figure 2 is a schematic diagram of a low-speed merging scene provided by an embodiment of the present application;

[0077] Figure 3 is a schematic diagram of a closed-loop vehicle decision control provided by an embodiment of the present application;

[0078] Figure 4 is a grid diagram of the trajectory termination discrete position in the action space trajectory planning provided by an embodiment of the present application;

[0079] Figure 5 is a decision process scene schematic diagram provided by an embodiment of the present application, wherein (a) is a scene schematic diagram at t=0, (b) is a scene schematic diagram at t=2s, (c) is a scene schematic diagram at t=4s, and (d) is a scene schematic diagram at t=6s. DETAILED DESCRIPTION

[0080] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0081] The application proposes a layered game decision method for autonomous vehicle in low-speed merging scene, which includes two stages of merging lane cut-in decision of upper layer finite state machine and trajectory planning of lower layer game theory in the layered game decision framework.

[0082] The merging lane cut-in decision stage of upper layer finite state machine: in low-speed merging scene, the goal of autonomous vehicle is to merge into target lane dynamically and safely, which can be divided into six discrete stages: preparation, vehicle gap determination, interactive game, lane change, emergency braking avoidance, and completion of merging based on finite state machine (FSM), and the transition between states is controlled by rules. First, the autonomous vehicle evaluates the surrounding vehicle state, selects the optimal safe vehicle gap, and determines the target vehicle for game theory trajectory planning; second, the autonomous vehicle (merging vehicle) and the target vehicle interact with each other, and a safe and feasible Nash equilibrium solution is obtained, if the game interaction fails (no feasible Nash equilibrium solution), the autonomous vehicle slows down to avoid the target vehicle; then, the autonomous vehicle executes the control instructions in the Nash equilibrium solution, and dynamically detects the collision risk in the lane change process, if the collision risk index exceeds the threshold during execution, it is also considered as game interaction failure of autonomous vehicle, and emergency braking is selected to terminate lane change to avoid the target vehicle; at the same time, when the game interaction fails, the autonomous vehicle game returns to the vehicle gap determination stage, this iterative vehicle gap detection strategy seeks new merging opportunities while maintaining a safe following distance; finally, the autonomous vehicle safely merges into the target vehicle and completes the merging task.

[0083] The trajectory planning stage of lower layer game theory: since there is potential conflict between the trajectories of autonomous vehicle (merging vehicle) and target vehicle, the interaction between the two vehicles is modeled as a game problem. First, according to the current state information of autonomous vehicle and target vehicle, feasible motion trajectories are planned for the two vehicles in the action space; second, according to the designed cost function, the cost of each vehicle for different combinations of trajectories is calculated , obtaining a double cost matrix; third, Nash equilibrium is found in the established cost matrix: in most cases, pure strategy equilibrium can be obtained by iteratively eliminating strictly dominated strategies, for other minority cases, the Lemke-Howson method can be used to solve the mixed strategy in the game; fourth, safety protection check is performed on the obtained optimal strategy to avoid possible conflicts, the future state of autonomous vehicle and vehicles on the target lane is predicted through the kinematics equation of the vehicle and the action sequence in the optimal strategy; finally, if there is a collision in the predicted trajectory of the vehicle, the autonomous vehicle selects emergency braking to terminate lane change to avoid the target vehicle; otherwise, the obtained Nash equilibrium is output as the current optimal strategy.

[0084] Further, the merging decision stage of the upper layer finite state machine divides the merging maneuver of the autonomous vehicle into six discrete stages, ensuring that the autonomous vehicle maintains stability while safely and efficiently merging. The specific implementation is that the six running stages of the vehicle merging finite state machine FSM are: the preparation stage: waiting for the merging instruction in the straight lane; the vehicle gap determination stage: selecting the optimal safe merging distance through risk assessment; the interactive game stage: negotiating the optimal trajectory with the rear target vehicle through Nash equilibrium solution; the lane changing stage: executing the lower layer output strategy and detecting the collision risk in real time; the emergency braking avoidance stage: decelerating to terminate lane changing to avoid the rear target vehicle straight through; and the completion of merging into the target lane.

[0085] The transition between the six running stages is controlled by rules, and the rule-based transition logic is as follows:

[0086] Preparation stage→vehicle gap determination stage: receiving the merging instruction of the intelligent driving assistance system or the driver before reaching the end of the road;

[0087] Vehicle gap determination stage→game interaction stage: selecting the optimal vehicle gap by minimizing the risk index The calculation formula is:

[0088] ;

[0089] Wherein, represents the longitudinal position coordinate of the autonomous vehicle, i.e., the merging vehicle, represents the longitudinal position coordinate of the vehicle in front of the gth vehicle gap, represents the longitudinal position coordinate of the vehicle behind the gth vehicle gap; , represents the minimum safe distance, which is set according to the actual situation and experience.

[0090] Game interaction stage→lane changing stage / emergency braking avoidance stage: triggering lane changing when there is a feasible Nash equilibrium; if the game fails, decelerate and give way;

[0091] Lane changing stage→completion stage / emergency braking avoidance stage: entering the completion stage after completing merging; if the risk exceeds the threshold, triggering emergency braking and giving way;

[0092] Emergency braking avoidance stage→vehicle gap determination stage: returning to vehicle gap search after the target straight vehicle passes, to select the next feasible gap, and reinitialize the game interaction with the newly assigned rear vehicle.

[0093] This finite state machine transition process selects the safety gap through repeated iterations, and finally the autonomous vehicle completes the merging operation, ensuring that the autonomous vehicle can merge safely and efficiently while maintaining operational stability in low-speed merging scenarios.

[0094] A hierarchical game-theoretic decision-making method for autonomous vehicles in low-speed merging scenarios, such as... Figure 1 As shown, it includes:

[0095] First, the autonomous vehicle receives a merging command and prepares to begin the merging phase; then, it detects the optimal merging vehicle gap from the feasible gaps in the target lane and marks the vehicles in front of and behind this merging gap as the preceding vehicle and the following vehicle, respectively.

[0096] Then, the interactive game phase begins. The interaction between the autonomous vehicle (merging vehicle) and the target vehicle (following vehicle) can be described as a game problem. Based on the current state information of the autonomous vehicle and the target vehicle, feasible trajectories are planned for both vehicles within the action space. Then, according to the designed cost function, the cost incurred by each vehicle for different combinations of trajectories is calculated. , get one The system first constructs a dual cost matrix. Then, it searches for a Nash equilibrium within the established cost matrix. In most cases, a pure policy equilibrium can be obtained by iteratively eliminating strictly dominant policies. For other rare cases, the Lemke-Hausen method can be used to solve mixed policies in the game. Next, the obtained Nash equilibrium is used as the current optimal policy output, and a safety protection check is performed on the obtained optimal policy to avoid possible conflicts. The future states of the autonomous vehicle and the vehicles in the target lane are predicted through the vehicle's kinematic equations and the action sequence in the optimal policy. Finally, if a collision is predicted in the vehicle's trajectory, the autonomous vehicle chooses to brake suddenly to terminate the lane change in order to avoid the straight-going target vehicle. Otherwise, the obtained Nash equilibrium is used as the current optimal policy output.

[0097] Next, the autonomous vehicle executes the control commands from the Nash equilibrium solution and dynamically detects collision risks in real time during the lane-changing process. If the collision risk index exceeds a threshold during execution, it is considered a failure of the autonomous vehicle's game interaction, and it chooses to brake suddenly to terminate the lane change and avoid the straight-going target vehicle. Simultaneously, when the game interaction fails, the autonomous vehicle returns to the vehicle gap determination stage. This iterative vehicle gap detection strategy maintains a safe following distance while searching for new merging opportunities. Finally, the autonomous vehicle safely merges into the target vehicle, completing the merging task. This closed-loop process iterates through the hierarchical game decision-making process, selecting feasible gaps one by one, until the merging into the target lane is completed safely. This ensures that the autonomous vehicle can make real-time decisions in low-speed merging scenarios.

[0098] The method of the embodiment addresses the problem that traditional automatic driving decision models are often based on preset and conservative rules and are difficult to cope with the uncertain interaction intention of target lane vehicles in merging scenarios. A low-speed merging scenario automatic driving hierarchical game decision method considering collision risk is proposed. A hierarchical game decision framework combining the lane merging decision stage of the upper layer finite state machine and the trajectory planning stage of the lower layer game theory is constructed, and a closed-loop iteration mechanism is introduced to realize the dynamic decision-making capability of automatic driving vehicles in low-speed merging scenarios, which is efficient, safe, and consistent with actual traffic interaction. First, for low-speed merging scenarios, the merging process of the vehicle is divided into six stages. Second, the optimal vehicle gap is selected by minimizing the risk index. Third, the automatic driving vehicle (merging vehicle) and the target vehicle interact in a game to obtain a safe and feasible Nash equilibrium solution. If the game interaction fails (no feasible Nash equilibrium solution), the automatic driving vehicle slows down to avoid the straight-ahead target vehicle. Then, the automatic driving vehicle executes the control instructions in the Nash equilibrium solution and dynamically detects the collision risk during the lane changing process. If the collision risk index exceeds the threshold during execution, the automatic driving vehicle is also considered to have failed in game interaction, and emergency braking is selected to terminate lane changing to avoid the straight-ahead target vehicle, which effectively avoids potential risks caused by model uncertainty or abnormal interaction, greatly improving the robustness and safety of the system. At the same time, when the game interaction fails, the automatic driving vehicle returns to the game distance selection stage. This iterative vehicle gap detection strategy maintains a safe following distance while seeking new merging opportunities, enabling real-time switching of the gap and reinitializing the game process when no feasible solution is obtained, thereby ensuring real-time response and decision reliability of the system in continuously changing scenarios.

[0099] In the lower layer game theory trajectory planning stage of the method of the embodiment, the interaction between the automatic driving vehicle (merging vehicle) and the target lane front and rear vehicles is modeled as a game problem. Specifically, first, the automatic driving vehicle (merging vehicle) AV and the rear vehicle SV are determined as the two participants in the game. Based on the current state (position, speed, and acceleration) of the vehicle, the vehicle motion trajectory within the planning time is planned based on a 5th order polynomial , and the action space of the merging vehicle and the rear vehicle is obtained, respectively.

[0100] As shown in Figure 2 , for a typical strong interaction traffic scenario, a dynamic game model is established based on the three elements of the game (participants, action space, and strategy cost), where represents the participants in the game, represents the action space of the game, represents the cost function of the participants, and the automatic driving vehicle AV and the target vehicle SV are determined as the participants in the game The implementation of planning a feasible motion trajectory in the action space is: setting a certain time A set of sampled vehicle motion trajectories are planned as the action space of the vehicle, considering that the lateral motion of the front vehicle and the lane-changing decision of the merging vehicle have less influence, only the longitudinal motion of the front vehicle is modeled, and the lateral and longitudinal motions of the merging vehicle and the rear vehicle are modeled based on a 5th order polynomial:

[0101]

[0102] wherein, represents the vehicle trajectory, i.e., the position coordinates of the vehicle in the longitudinal and lateral directions changing with time, represents the coefficients of the polynomial equation, is determined by the initial state T init = [ x init , y init , x ˙ init , y ˙ init , x ¨ init , y ¨ init ] and the terminal state T term = [ x term , y term , x ˙ term , y ˙ term , x ¨ term , y ¨ term ] of the vehicle trajectory, wherein, respectively represent the positions of the vehicle in the longitudinal and lateral directions, respectively represent the velocities of the vehicle in the longitudinal and lateral directions, respectively represent the accelerations of the vehicle in the longitudinal and lateral directions. Assuming that the time length of the vehicle trajectory is (i.e., the time value is ), the and are brought into and its first derivative and second derivative , the coefficients of the polynomial equation can be solved. The initial state is the current state of the vehicle, and it is assumed that the lateral motion of the vehicle at the terminal time has been completed, and the control input to the vehicle is zero, i.e., it satisfies Therefore, by changing only the vehicle position and the longitudinal velocity at the terminal time, a unique vehicle trajectory can be determined.

[0103] As shown in Figure 3 , in order to balance the planning accuracy and real-time calculation complexity, the vehicle position at the terminal time is discretized within the feasible region, and the planning resolution of the merging vehicle AV and the rear vehicle SV is , wherein respectively represent the interval of the discrete position of vehicle j in the longitudinal and lateral directions at the terminal time; the longitudinal velocity of the vehicle at the terminal time is discretized as , wherein ​​The number of discrete longitudinal velocity, discrete interval is 1; and the number of feasible motion trajectory of two vehicles (merging vehicle and rear vehicle) in the action space is obtained: , wherein and respectively represent the discrete number of vehicle position at the end of longitudinal and transverse directions; and the motion trajectory in the action space is filtered based on the scene and kinematic constraints, and the action space of the game is obtained A j = [ T 1 j , T 2 j ,..., T M j j ] The kinematic constraints are vehicle parameters, including longitudinal and transverse coordinates, maximum longitudinal and transverse velocities, maximum longitudinal and transverse accelerations, etc.

[0104] Further, considering safety, efficiency and comfort, the cost function of each participant is designed , , so as to obtain the game decision model of the interaction of the two vehicles. The running cost of each vehicle in the game model at the kth time step is designed:

[0105] ;

[0106] , wherein respectively represent the running cost of the two vehicles (merging vehicle and rear vehicle) from the kth time step to the (k+1)th time step, respectively represent the state of the merging vehicle, the front vehicle and the rear vehicle at the kth time step, the state includes longitudinal and transverse positions and velocities s k j = [ x k j , y k j , x ˙ k j , y ˙ k j ] T , , respectively represent the control input of the selected trajectory of the two vehicles in the action space, the control input is the longitudinal and transverse acceleration u k j = [ x ¨ k j , y ¨ k j ] T , , r k AV = [ r 1 AV , r 2 AV , r 3 AV , r 4 AV ] and r k SV = [ r 1 SV , r 2 SV , r 3 SV , r 4 SV , r 5 SV ] respectively represent the combination of each part of the cost function of the two vehicles, ω AV = [ w 1 AV , w 2 AV , w 3 AV , w 4 AV ] T and ω SV = [ w 1 SV , w 2 SV , w 3 SV , w 4 SV , w 5 SV ] T respectively represent the weight vector of the cost function of the two vehicles, and the weight vector is flexibly adjusted according to the driving style preference of the vehicle.

[0107] Considering safety, efficiency and comfort, the components of the cost function of the merging vehicle are defined as:

[0108] represents the collision safety component of the merging vehicle, that is, if any collision occurs among the merging vehicle, the front vehicle and the rear vehicle, , otherwise The weight of the collision safety component is greater than the weight of the efficiency and comfort cost components, to primarily ensure safety.

[0109] represents the safety constraint component of the merging vehicle, i.e. the merging vehicle should keep a safe distance from the end of the lane where it is located:

[0110] ;

[0111] wherein, represents the longitudinal distance of the merging vehicle from the end of the lane where it is located, represents the lateral position coordinate of the merging vehicle, represents the minimum longitudinal safety distance of the merging vehicle from the end of the lane where it is located, which is set according to the scenario and experience;

[0112] represents the efficiency component of the merging vehicle, i.e. the merging vehicle should complete the lane-changing task as soon as possible, which is defined as:

[0113] ;

[0114] wherein, represents the lane width of the road;

[0115] represents the comfort component of the merging vehicle, which is defined as:

[0116] ;

[0117] wherein, and respectively represent the acceleration values of the merging vehicle in the longitudinal and lateral directions, and respectively represent the maximum acceleration values of the merging vehicle in the longitudinal and lateral directions;

[0118] Similarly, the components of the cost function of the rear vehicle are defined as:

[0119] represents the collision safety component of the rear vehicle, i.e. if any collision occurs between the merging vehicle, the front vehicle and the rear vehicle, , otherwise ; the weight of the collision safety component is greater than the weight of the efficiency and comfort cost components, to primarily ensure safety;

[0120] represents the safety constraint component of the rear vehicle, i.e. the rear vehicle should keep a safe distance from the front vehicle:

[0121] ;

[0122] wherein, denotes the longitudinal distance between the rear vehicle and the front vehicle, where denote the lateral position coordinate values of the front vehicle and the rear vehicle, respectively, denotes the minimum longitudinal safety distance between the rear vehicle and the front vehicle, which is set according to the scene and experience;

[0123] denotes the efficiency component of the rear vehicle, i.e., the rear vehicle should pass through the merging area as soon as possible, which is defined as:

[0124] ;

[0125] wherein, denotes the longitudinal distance between the rear vehicle and the end of the merging area;

[0126] denotes the ride comfort component of the rear vehicle, which is defined as:

[0127] ;

[0128] wherein, and denote the acceleration values of the rear vehicle in the longitudinal and lateral directions, respectively, and denote the maximum acceleration values of the rear vehicle in the longitudinal and lateral directions, respectively;

[0129] wherein, denotes the penalty term for the rear vehicle deviating from the lane centerline, which is defined as: wherein, denotes the lateral position coordinate of the rear vehicle;

[0130] Based on the definitions of the components of the cost function, the function weights can be adjusted according to different driver types , and the specific adjustment method is as follows: if the driving style of vehicle j is aggressive, it means that the ride comfort is more preferred to efficiency, i.e., the design parameters satisfy ; on the contrary, if the driving style of vehicle j is conservative, it means that the ride comfort is more preferred to safety and stability, i.e., the design parameters satisfy . For the trajectory planned in the action space, the total cost of completing the trajectory tracking task is which is obtained by accumulating the running costs of multiple steps:

[0131] ;

[0132] wherein, and denote the discrete time steps required for the merging vehicle and the rear vehicle to complete the task according to the trajectory .

[0133] Furthermore, after defining the action space and cost function, we obtain a The dual cost matrix is ​​obtained, and then the Nash equilibrium is found in the established cost matrix. The specific implementation method is as follows:

[0134] According to the definition of Nash equilibrium: any strategy adopted by a player in a game is the optimal response strategy among all other players' strategies. Mathematically, if a certain strategy combination... (Composed of a strategy chosen by each player in the game) satisfies: for any player j, their strategy It is the optimal response, i.e., the inequality. If this holds true, the combination is called a Nash equilibrium.

[0135] In this invention, an iterative elimination of strictly disadvantaged strategies is first employed to find a pure strategy equilibrium. If no pure strategy equilibrium can be found, the Lemke-Hausen method is then used to find mixed strategies in the game. A mixed strategy in a game means that participants choose actions in the action space according to a pre-defined probability distribution, which is defined as a combination of probability distributions. :

[0136] ;

[0137] In the formula, , These represent the selected motion trajectory of vehicle j. The probability distribution, p i j ∈ [ 0 , 1 ] , i ∈ { 1 ,..., M j } , j ∈ { AV , SV } This indicates that vehicle j selects a motion trajectory. The probability distribution is obtained using the Lemke-Hausen method.

[0138] Furthermore, after finding the optimal Nash equilibrium solution, a security protection check is performed on the obtained optimal strategy to avoid potential conflicts. The specific implementation method is as follows:

[0139] For autonomous vehicles and vehicles in the target lane, the future state of the vehicle is predicted using the vehicle's kinematic equations and the action sequence in the optimal policy:

[0140] ;

[0141] In the formula, k represents the current time. Indicates that vehicle j is in The internal state, This represents the optimal strategy after the upper-level game. Indicates the optimal policy π * = [ T AV ,* , T SV ,* ] and current state The obtained future state of the vehicle is obtained through trajectory equations. and its first derivative and its second derivative , respectively, denote the motion trajectory of AV and SV in the optimal strategy, , , respectively, denote the motion trajectory of AV and SV in the optimal strategy, denotes the predicted time length. Then, the predicted trajectory of the vehicle is checked for collision: if there is a collision in the predicted trajectory of the vehicle, the AV performs emergency braking to avoid the vehicle on the target lane, i.e. denotes the emergency braking control amount of the vehicle), otherwise, the vehicle executes the optimal strategy obtained in the upper dynamic feedback decision-making stage, i.e. to complete the merging task, Figure 4 a closed-loop vehicle decision-making control schematic diagram is given.

[0142] Further, in the lane-changing (stage 4) of the merging decision of the upper finite state machine of the method, the AV executes the control instructions in the Nash equilibrium solution while performing real-time dynamic detection of the collision risk in the lane-changing process, and the specific implementation manner is as follows: based on two indexes (the modified collision time and the modified headway) and the softmax function, a comprehensive merging risk index (CRI) in the lane-changing process is designed:

[0143] ;

[0144] In the formula, and respectively denote the Sigmoid regularization index of the modified collision time (MTTC) and the modified headway (MTHW):

[0145] ;

[0146] In the formula, are all sensitivity parameters used to adjust the collision risk. MTTC represents the modified collision time, which is used to evaluate the collision risk of the AV and the rear vehicle, and MTHW represents the modified headway, which is used to evaluate the driving risk of the rear vehicle:

[0147] ;

[0148] ;

[0149] In the formula, and respectively denote the distance of the AV and the rear vehicle to the conflict area, and respectively denote the speed of the AV and the rear vehicle, denotes that the AV has entered the target lane. ​

[0150] Finally, if the game interaction in stage 3 fails (no feasible Nash equilibrium solution) or the interaction fails when performing lane changing in stage 4 (risk exceeds the threshold), the autonomous vehicle needs to enter the emergency braking avoidance stage (stage 5), and then return to the inter-vehicle gap selection stage to perform the avoidance operation (yielding to the "rear vehicle") to select the next feasible gap and reinitialize the game with the newly assigned rear vehicle. This closed-loop iteration mechanism can ensure that the autonomous vehicle successfully merges into the target lane. Figure 5 A scene diagram of an example of a decision-making process is given, when t = 0 (as shown in Figure 5 (a)), the autonomous vehicle (merging vehicle) selects a target gap, and through the hierarchical decision-making framework, obtains a safe and feasible trajectory to execute; when t = 2 seconds (as shown in Figure 5 (b)), the merging vehicle begins to turn to prepare for lane changing, while the rear vehicle slightly turns right and slows down to reduce the risk of collision; when t = 4 seconds (as shown in Figure 5 (c)), the merging vehicle completely enters the target lane, and the rear vehicle begins to slightly turn left to correct the vehicle running angle; when t = 6 seconds (as shown in Figure 5 (d)), the merging vehicle completes the task of merging into the target lane. This mechanism greatly enhances the flexibility and adaptability of the system in responding to unexpected situations, ensuring that the autonomous vehicle can eventually find a safe merging opportunity in complex real traffic flow, thereby significantly improving the success rate of the merging task and the overall fault tolerance of the system.

[0151] Overall, the embodiment of the present application achieves significant comprehensive technical effects through its hierarchical game decision-making framework and closed-loop iteration mechanism. The method decouples and integrates intelligent decision-making and safety protection: the lane merging decision-making stage of the upper layer finite state machine decomposes multiple states of the merging process of the vehicle and designs state transition rules. This merging strategy considering collision risk enables the autonomous vehicle to simulate the interactive intention of human drivers, generate a humanized and highly efficient negotiation smooth merging strategy, and significantly improve the rationality of the decision and the fusion degree of the traffic flow. The trajectory planning stage of the lower layer game theory serves as a system solution to the upper game problem by establishing a game model and solving the Nash equilibrium, and performs safety protection confirmation, which ensures the absolute safety of the system in any unexpected situation and enhances the robustness of the scheme. In addition, through limited action space planning and preferential use of efficient equilibrium solving algorithms, the real-time performance of the decision is guaranteed, meeting the computational power constraints of the vehicle platform; and the closed-loop iteration mechanism gives the system the ability to continuously optimize and actively correct errors, and when the strategy fails, it can safely transition and re-game until a collision-free trajectory is generated, thereby greatly improving the success rate of the task and environmental adaptability in complex low-speed merging scenarios.

[0152] The embodiment considers that the automatic driving vehicle interacts with other vehicles around and makes optimal decisions in real time, and combines safety, efficiency and comfort and other traffic demands, constructs a hierarchical game decision framework combined with a closed-loop iteration mechanism, and realizes the real-time decision function of the automatic driving vehicle in the low-speed merging scene.

[0153] The application further provides an electronic device including a memory and a processor, the memory storing a computer program, and the processor implementing the steps of the method when executing the computer program.

[0154] The electronic device can be a desktop computer, a notebook computer, a palm computer, a cloud server and the like. The processor can be a central processing unit (CPU), and can also be other general-purpose processors, digital signal processors (DSP), application specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components and the like. The memory can be used to store computer programs and / or modules, and the processor can run or execute the computer programs and / or modules stored in the memory, and call data stored in the memory, so as to realize various functions of the electronic device.

[0155] The application further provides a computer readable storage medium, which stores a computer program, and the computer program is executed by a processor to implement the steps of the method.

[0156] Specifically, the memory can include a high-speed random access memory, and can also include a non-volatile memory, for example, a hard disk, a memory, a plug-in hard disk, a smart media card (SMC), a secure digital (SD) card, a flash card, at least one disk storage device, a flash memory device, or other volatile solid-state storage devices.

[0157] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the application, and not to limit them; although the application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for part or all of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the application.

Claims

1. A hierarchical game-theoretic decision-making method for autonomous vehicles in low-speed merging scenarios, characterized in that, The method includes: Step S1: The autonomous vehicle receives the inbound instruction and prepares to begin the inbound phase. Step S2, vehicle gap determination: The autonomous vehicle assesses the status of surrounding vehicles, selects the safe and optimal vehicle gap, and determines the target vehicle for game theory trajectory planning. Step S3: The autonomous vehicle and the target vehicle engage in interactive game to obtain a safe and feasible Nash equilibrium solution. Step S4: The autonomous vehicle executes the control instructions in the Nash equilibrium solution and performs real-time dynamic detection of collision risks during the lane-changing process. Step S5: When the game interaction fails, the autonomous vehicle game returns to the vehicle gap determination stage, and seeks new merging opportunities while maintaining a safe following distance. Step S6: The autonomous vehicle safely merges into the target vehicle, completing the merging task; In step S2, the optimal vehicle gap is selected by minimizing the risk index. The calculation method is as follows: ； in, Represents the longitudinal position coordinates of the autonomous vehicle. This represents the longitudinal position coordinate of the vehicle ahead at the g-th vehicle gap. This represents the longitudinal position coordinates of the vehicle behind at the g-th vehicle gap. , Indicates the minimum safe distance; In step S3, the interaction between the autonomous vehicle and the target vehicle is described as a game theory problem. Based on the current state information of the autonomous vehicle and the target vehicle, feasible motion trajectories are planned for both vehicles in the action space. Then, according to the designed cost function, the cost incurred by each vehicle for different combinations of trajectories is calculated. , get one The system first obtains a dual cost matrix. Then, it searches for a Nash equilibrium within the established cost matrix. Next, it outputs the obtained Nash equilibrium as the current optimal policy and performs a safety protection check on the obtained optimal policy to avoid possible conflicts. It then predicts the future state of the autonomous vehicle and the vehicles in the target lane using the vehicle's kinematic equations and the action sequence in the optimal policy. Finally, if a collision is predicted in the vehicle's trajectory, the autonomous vehicle chooses to brake urgently to terminate the lane change in order to avoid the straight-going target vehicle. Otherwise, it outputs the obtained Nash equilibrium as the current optimal policy.

2. The method according to claim 1, characterized in that, Planning feasible motion trajectories within the action space specifically includes: Set at a certain time An internal plan is created using a set of sampled vehicle trajectories as the vehicle's motion space. The lateral and longitudinal motions of the two vehicles are modeled based on a 5th-order polynomial: ； in, This represents the vehicle's trajectory, specifically its position coordinates in the longitudinal and lateral directions over time. Denotes the coefficients of a polynomial equation. , From the initial state of the vehicle trajectory and termination state It is confirmed that, among them, These indicate the vehicle's position in the longitudinal and lateral directions, respectively. These represent the vehicle's speed in the longitudinal and lateral directions, respectively. These represent the vehicle's acceleration in the longitudinal and lateral directions, respectively. Discretize the vehicle position at the termination time within the feasible region, and the planning resolution for merging the vehicle AV and the following vehicle SV is: ,in Let represent the intervals of the discrete positions of vehicle j in the longitudinal and lateral directions at the termination time, respectively. Then, the longitudinal velocity of the vehicle at the termination time is discretized as... ,in This represents the number of discrete longitudinal velocities, with a discrete interval of 1, yielding feasible motion trajectories for the merging vehicle and the following vehicle within the action space. Its discrete quantity is: ,in, and These represent the number of discretized vehicle positions at the termination time in the longitudinal and lateral directions, respectively. Based on scene and kinematic constraints, the planned motion trajectories in the action space are filtered out, that is, all motion trajectories that satisfy the upper and lower limits of vehicle state variables in the scene are selected as feasible action spaces. The upper and lower limits of vehicle state variables include the vehicle's lateral and longitudinal coordinates, maximum longitudinal and lateral velocities, and maximum longitudinal and lateral accelerations.

3. The method according to claim 2, characterized in that, Identify the inbound vehicle AV and the target vehicle SV as the participants in the game, and design the vehicle running cost for each vehicle in the established game model at time step k: ； in, Let represent the travel costs of the merging vehicle and the following vehicle from time k to time k+1, respectively. These represent the states of the merging vehicle, the preceding vehicle FV, and the following vehicle SV at time k, respectively. The states include their longitudinal and lateral positions and velocities. , , These represent the control inputs for the selected trajectories of the two vehicles in the motion space, namely the longitudinal and lateral accelerations. , , and These represent the combinations of the various indices of the cost functions for the two vehicles. and These represent the cost function weight vectors for the two vehicles, which are flexibly adjusted according to the driving style preferences of the vehicles. The cost function metrics for imported vehicles are defined as follows: This represents the collision safety component for merging vehicles, i.e., the component that would be safe if any collision were to occur between the merging vehicle, the vehicle in front, and the vehicle behind. ,otherwise The weight of the collision safety component is greater than the weight of the efficiency and comfort cost components; This represents the safety constraint component for merging vehicles, specifically the safe distance that merging vehicles should maintain from the end of their lane. ； in, This indicates the longitudinal distance of merging vehicles from the end of their respective lanes. This indicates the lateral position coordinates of the merging vehicle. This indicates the minimum longitudinal safety distance between merging vehicles and the end of their lane; The efficiency component representing the input vehicle is defined as follows: ； in, Indicates the width of the road lanes; The ride comfort component of the vehicle is represented and defined as: ； in, and These represent the longitudinal and lateral acceleration values ​​of the merging vehicle, respectively. and These represent the maximum acceleration values ​​of the merging vehicle in the longitudinal and lateral directions, respectively; Similarly, the cost function metrics for the following vehicle are defined as follows: This represents the collision safety component for the vehicle behind, i.e., if any collision occurs between the merging vehicle, the vehicle in front, and the vehicle behind, then... ,otherwise The weight of the collision safety component is greater than the weight of the efficiency and comfort cost components; This represents the safety constraint component for the following vehicle, meaning the following vehicle should maintain a safe distance from the vehicle in front: ； in, This indicates the longitudinal distance between the following vehicle and the vehicle in front. These represent the lateral position coordinates of the front and rear vehicles, respectively. This indicates the minimum longitudinal safe distance between the following vehicle and the vehicle in front; The efficiency component of the following vehicle is defined as: ； in, This indicates the longitudinal distance between the following vehicle and the end of the merging area; The component representing the ride comfort of the following vehicle is defined as: ； in, and These represent the longitudinal and lateral acceleration values ​​of the following vehicle, respectively. and These represent the maximum acceleration values ​​of the following vehicle in the longitudinal and lateral directions, respectively; The penalty for a vehicle following a vehicle deviating from the center line is defined as follows: ,in, Indicates the lateral position coordinates of the following vehicle; For a planned trajectory in the action space, the total cost of completing the trajectory tracking task is... The results are derived by accumulating the running costs of multiple steps: ； in, and These represent the discrete time steps required for the merging vehicle and the following vehicle to complete the task according to the trajectory, respectively. Finally, based on the total cost, we obtain a... The dual cost matrix.

4. The method according to claim 3, characterized in that, Find the Nash equilibrium in the established cost matrix, including: First, an iterative approach is used to eliminate strictly disadvantaged strategies to find a pure strategy equilibrium. If no pure strategy equilibrium can be found, the Lemke-Hausen method is then used to find mixed strategies in the game, defining a probability distribution combination. : ； in, , This indicates that vehicle j selects a motion trajectory. The probability distribution, This indicates that vehicle j selects a motion trajectory. The probability distribution is obtained using the Lemke-Hausen method.

5. The method according to claim 1, characterized in that, Step S4 involves real-time dynamic detection of collision risks during lane changing, including: The Comprehensive Lane Change Risk Index (CRI) is designed based on two indicators: collision time correction, headway correction, and a softmax function. ； in, and The Sigmoid regularization indices representing corrected time to collision (MTTC) and corrected headway (MTHW) are respectively: ； in, Both are sensitivity parameters used to adjust collision risk. MTTC represents corrected collision time, and MTHW represents corrected headway. ； ； in, and These represent the distances from the autonomous vehicle and the following vehicle to the conflict zone, respectively. and These represent the speeds of the autonomous vehicle and the vehicle behind it, respectively. This indicates that the autonomous vehicle has entered the target lane.

Images