Mixed traffic flow human-like confluence trajectory control method based on random potential game

Abstract

The invention discloses a random potential game-based mixed traffic flow human-like confluence trajectory control method, which belongs to the technical field of automatic driving, and comprises the following steps of: establishing a random potential game model considering communication delay or sensing uncertainty influence, determining a potential function in the random potential game model, and determining a random potential function in the random potential game model; calibrating parameters in the random potential game model by using an open source data set; collecting real-time state information of all vehicles in a highway confluence area range as input information of trajectory planning; based on the random potential game model, a model prediction controller is constructed, and a distributed algorithm is adopted for solving; and calculating a vehicle action combination at the next moment through the model by using the real-time state information of the vehicle to obtain a vehicle planning track, and updating the state of the vehicle. According to the invention, the driving decision control logic of the human driver can be accurately described and the planning trajectory can be generated, so that the automatic driving vehicle and the manual driving vehicle realize good interaction, and the overall traffic efficiency and safety of the confluence area are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of autonomous driving, and particularly relates to a human-like merging trajectory control method for mixed traffic flow based on stochastic potential game. Background Art

[0002] The highway ramp merging area is a typical traffic bottleneck area, and potential conflicts are often caused by the merging of ramp vehicles. With the development of vehicle networking and autonomous driving technologies, vehicles can achieve microscopic trajectory control through vehicle networking technologies, thereby effectively eliminating the above conflicts and significantly improving the traffic efficiency and safety of the merging area. Currently, many studies focus on cooperative merging strategies in a pure autonomous driving environment, aiming to optimize the interaction and coordination between autonomous vehicles. These studies have made significant progress in improving the cooperative performance of autonomous vehicles, laying a solid foundation for the development of autonomous driving technologies.

[0003] Despite the significant development of autonomous driving technologies, there are still many challenges in their practical applications in a mixed traffic flow environment (coexistence of autonomous and human-driven vehicles). Firstly, existing research pays less attention to merging control in a mixed flow environment and is difficult to effectively handle the interaction problems between autonomous and human-driven vehicles. Secondly, the research on human-like interaction mechanisms is limited, especially how different driving styles affect merging decisions and control has not been fully explored. In addition, most existing decentralized control algorithms only optimize the behavior of agents from an individual perspective and rarely consider the impact of individual behavior on the global system, resulting in insufficient overall coordination. Finally, existing research often ignores the uncertainties caused by sensor errors or communication delays, and these problems are difficult to handle in practical applications, further reducing the interaction level between autonomous and human-driven vehicles and limiting the wide application of autonomous driving technologies in a mixed traffic flow environment. Summary of the Invention

[0004] To solve the above technical problems, the present invention proposes a human-like merging trajectory control method for mixed traffic flow based on stochastic potential game to solve the problems existing in the above prior art.

[0005] To achieve the above object, in a first aspect, the present invention provides a human-like merging trajectory control method for mixed traffic flow based on stochastic potential game, including:

[0006] Establish a stochastic potential game model considering the influence of communication delay or sensing uncertainty, determine the potential function in the stochastic potential game model, and calibrate the parameters in the stochastic potential game model using an open-source dataset;

[0007] Collect the real-time state information of all vehicles within the scope of the highway merging area as the input information for trajectory planning;

[0008] Based on the stochastic potential game model, a model predictive controller is constructed and solved using a distributed algorithm; using the real-time vehicle state information, the vehicle action combination at the next moment is calculated through the stochastic potential game model to obtain the vehicle planned trajectory and update the vehicle state.

[0009] Preferably, the process of establishing a stochastic potential game model considering the influence of communication delay or sensing uncertainty includes:

[0010] According to the vehicle finding the best action under the stochastic potential game model, the original objective is converted into a mathematical expectation with a random vector, and the deterministic constraint is converted into a chance constraint;

[0011] The sampling average approximation method is used to convert the uncertain mathematical expectation and conditional constraints into a deterministic problem;

[0012] Determine the potential function in the game model, and the potential function is a continuous and exact potential game potential function;

[0013] Extract the vehicle trajectories with interactions in the highway merge area from the open-source dataset and calibrate the parameters using the model method.

[0014] Preferably, the potential function includes self-correlation cost and coupling cost. The self-correlation cost includes the passing efficiency cost, driving comfort cost, and cost of deviating from the target state, and the coupling cost includes the safety cost.

[0015] Preferably, the parameter calibration uses a bilevel programming model. The upper-level model minimizes the mean absolute error between the predicted action and the real action, and the lower-level model solves the potential game equilibrium.

[0016] Preferably, all the real-time vehicle state information includes:

[0017] Collect the current longitudinal and lateral positions, longitudinal and lateral speeds, and longitudinal and lateral accelerations of the vehicle as dynamic input information;

[0018] Collect the vehicle type, the lane to which the vehicle belongs, and the vehicle driving style factor as static input information;

[0019] Collect the merge area length, road width, and road boundary information.

[0020] Preferably, the steps of constructing the model predictive controller include:

[0021] Establish an MPC optimization model considering vehicle dynamics constraints, road conditions, traffic rules, and the behaviors of other vehicles;

[0022] Decompose the original optimization problem into a vehicle single-sided decision problem and an auxiliary variable update problem;

[0023] Solve it using a distributed algorithm.

[0024] Preferably, the steps of solving by using a distributed algorithm include:

[0025] When fixing the auxiliary variables and the dual variables, independently solve the unilateral decision problem for each vehicle to optimize its control strategy;

[0026] When fixing the vehicle control variables and the dual variables, coordinate the decisions among vehicles by solving the update problem of the auxiliary variables;

[0027] Adjust the dual variables according to the update rule of the Lagrange multiplier method to gradually satisfy the constraint conditions.

[0028] Preferably, the trajectory planning is a dynamic process, and generate the trajectories of all vehicles at the next moment according to the state at the current moment.

[0029] In a second aspect, the present invention also discloses a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0030] In a third aspect, the present invention also discloses a computer program product, including a computer program, and when the computer program is executed by a processor, the steps of the method described in the first aspect are implemented.

[0031] Compared with the prior art, the present invention has the following advantages and technical effects:

[0032] The present invention provides a human-like merging trajectory control method for mixed traffic flow based on stochastic potential game, including: First, establish a stochastic potential game model considering the influence of communication delay or sensing uncertainty, determine the potential function in the stochastic potential game model, and calibrate the parameters in the stochastic potential game model by using an open-source data set; Second, collect the real-time state information of all vehicles within the range of the highway merging area as the input information for trajectory planning; Finally, based on the stochastic potential game model, construct a model predictive controller and solve it by using a distributed algorithm; Use the real-time state information of the vehicle, calculate the vehicle action combination at the next moment through the stochastic potential game model, obtain the vehicle planned trajectory, and update the vehicle state.

[0033] The present invention considers the stochastic uncertainty caused by communication or sensing errors and stably and efficiently solves the optimal control of autonomous vehicles.

[0034] The present invention constructs a potential function to capture the decision-making logic and behavior randomness of human drivers; by understanding and learning the mechanism of human driving behavior, provide real-time human driving behavior prediction for autonomous vehicles in the highway merging area.

[0035] The present invention analyzes the driving behavior characteristics of different driving styles in natural driving data, calibrates the weight parameters in the stochastic potential game model, and depicts the decisions and reactions of human drivers under different traffic scenario conditions.

[0036] The present invention establishes a model predictive controller, and based on the calibrated game model, realizes human-like merging trajectory planning; adopts the model predictive control algorithm to solve the optimal control output and plan the future movement of the autonomous vehicle.

[0037] The present invention constructs a distributed solution algorithm, converts the centralized optimization into a distributed optimization framework, realizes the decentralized control of individual autonomous vehicles, and improves the solution efficiency. Finally, a good interaction between autonomous vehicles and human-driven vehicles is achieved, and the traffic operation efficiency and safety in the merging area are improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0038] The drawings constituting a part of this application are used to provide a further understanding of this application. The schematic embodiments of this application and their descriptions are used to explain this application and do not constitute an improper limitation to this application. In the drawings:

[0039] Figure 1 is the flowchart of the method according to the embodiment of the present invention;

[0040] Figure 2 is the schematic diagram of the various utilities changing with acceleration according to the embodiment of the present invention;

[0041] Figure 3 is the flowchart of parameter calibration according to the embodiment of the present invention;

[0042] Figure 4 is the generated planned trajectory and motion state diagram according to the embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0043] It should be noted that, without conflict, the embodiments in this application and the features in the embodiments can be combined with each other. The following will refer to the drawings and combine the embodiments to detail this application.

[0044] It should be noted that the steps shown in the flowchart of the drawings can be executed in a computer system such as a set of computer-executable instructions, and although the logical order is shown in the flowchart, in some cases, the steps shown or described can be executed in a different order than here.

[0045] Embodiment 1

[0046] As Figure 1 shown, the present embodiment provides a human-like merging trajectory control method for mixed traffic flow based on a stochastic potential game, including:

[0047] S1. Build a stochastic potential game model, considering the uncertainties caused by communication delays or sensing; determine the potential function in the model; calibrate the parameters in the model using an open-source dataset.

[0048] Specifically, S1 includes:

[0049] S11. By characterizing the mechanism of human driving behavior, establish a stochastic potential game model and find the optimal action:

[0050] (1)

[0051] (2)

[0052] In the formula, represents the optimal action, represents a random vector, indicating the uncertainties caused by communication delays or sensing. represents the mathematical expectation of. represents the probability of. The original objective is converted into the mathematical expectation with the random vector The deterministic constraint . is converted into a chance constraint , represents the failure probability.

[0053] S12. Use the sample average approximation method to make a deterministic transformation of the uncertain mathematical expectation and conditional constraints in step S11.

[0054] (3)

[0055] (4)

[0056] In the formula, represents the mathematical expectation of the cost, represents the th sample independently drawn from the vehicle initial state distribution. represents the maximum number of approximate samplings.

[0057] Specifically, the sample average approximation method is implemented as follows:

[0058] First, generate N samples from the random variable . Then, use the generated samples to construct an approximate deterministic optimization problem. For the objective function , its expected value can be approximately replaced by the average value of the samples:

[0059] (5)

[0060] Wherein, is the objective function of the sampling average approximation.

[0061] Then, solve the above deterministic optimization problem to obtain an approximate optimal solution .

[0062] Finally, evaluate the quality of the approximate solution. If the final solution does not meet the conditions, the number of samples N can be increased and the above steps can be repeated to improve the accuracy of the approximate solution. The sampling average approximation method can gradually improve the accuracy of the approximate solution by increasing the number of samples.

[0063] S13. The potential function in the game model needs to be a continuous and exact potential game potential function if and only if is a connected set, and there exists a potential function such that its partial derivative is consistent with the partial derivative of the cost function of each participant .

[0064] (6)

[0065] The potential function in the game model:

[0066] (7)

[0067] Wherein, represents the total potential function, and are the driving style weight factors, and represent the autocorrelation cost and the coupling cost respectively.

[0068] Specifically, the potential function can be expressed as:

[0069] (8)

[0070] (9)

[0071] (10)

[0072] (11)

[0073] (12)

[0074] Wherein, represents the weight factor of the human driving style, which can reflect the trade-off of the driver among different indicators. For example, an aggressive driver may pay more attention to traffic efficiency and ignore safety. Denotes the passing efficiency cost, which can be calculated by the sum of the squares of the deviations of the lateral and longitudinal speeds of the vehicle from the expected speed. If the longitudinal or lateral speed is closer to the expected speed, this cost will decrease. and Denotes the vehicle expected longitudinal and lateral speeds. Denotes the driving comfort cost, which can be calculated by the sum of the squares of the longitudinal and lateral accelerations of the vehicle. When the longitudinal or lateral acceleration decreases, this cost will decrease. Denotes the cost of deviating from the target state, including three parts: the square of the deviation of the lateral position from the center line of the target lane, the square of the lateral speed, and the square of the lateral acceleration. The target state is that the vehicle is driving on the center line of the target lane while its lateral speed and lateral acceleration are zero. Denotes the safety cost. If the and longitudinal or lateral spacing between the vehicle and the . Otherwise, , the safety cost will gradually increase as the vehicle spacing decreases, thus prompting the vehicle to stay away from surrounding vehicles. Among them, and represent the longitudinal and lateral safety distances respectively, and represent the length and width of the vehicle. The variation of utility with acceleration is as shown in Figure 2 .

[0075] S14. Extract the trajectories of vehicles with interactions in the highway merging area from the open-source dataset for a period of time, and use the following model method for parameter calibration:

[0076] (13)

[0077] (14)

[0078] In the formula, Denotes the optimal action of the potential game under the given parameter vector , and its result is the predicted value of the optimal action. Denotes the true acceleration.

[0079] Since is a non-linear programming problem, a heuristic algorithm can be selected to solve the optimal parameter vector in the upper-layer model. The lower-layer model is in the standard form of mathematical optimization, so global optimization can be selected for solution.

[0080] The algorithm flow chart of the parameter calibration framework is as shown in Figure 3 . Set the initial parameter vector , the number of iterations is 0; calculate the parameter vector of the cost function; real trajectory data can be extracted from the open-source dataset INTERACTION; solve the bilevel model, where the upper-level model is to minimize the mean absolute error between the predicted action and the real action, and the lower-level model is to solve the potential game equilibrium; if the preset iteration stop condition is reached, output the optimal estimated parameters , otherwise continue the iteration.

[0081] S2. Collect the real-time status information of all vehicles within the highway merge area as the input information for trajectory planning;

[0082] Specifically, S2 includes:

[0083] Use roadside facilities to collect the real-time status information of all vehicles within the highway merge area, including the current horizontal and vertical positions, horizontal and vertical speeds, and horizontal and vertical accelerations, as the real-time dynamic input information for trajectory planning; real-time static input information such as vehicle type, the lane to which the vehicle belongs, and the vehicle driving style factor; road structure information such as the length of the merge area, road width, and road boundaries. This information can be transmitted to autonomous vehicles for decision-making control; for manually driven vehicles, only the information of the vehicles in front in the current lane and in front in the target lane is transmitted as the decision-making for following and lane-changing control.

[0084] S3. Build a model predictive controller and solve it using a distributed algorithm; calculate the vehicle action combination at the next moment based on the real-time vehicle status information, obtain the vehicle planned trajectory, and update the vehicle status.

[0085] Specifically, the said S3 includes:

[0086] S31. The model predictive controller based on the stochastic potential game and solved with a distributed algorithm is:

[0087] (15)

[0088] (16)

[0089] In the formula, is an auxiliary variable.

[0090] The model predictive controller based on the stochastic potential game considers factors such as the vehicle dynamics constraints, road conditions, traffic rules, and the behaviors of other vehicles, and calculates the optimal control quantity in real time. A corresponding MPC optimization model can be established:

[0091] (17)

[0092] Among them, ; ; ; ;

[0093] In the formula, represents the cumulative cost of the potential function within the time range from to . . , , , are respectively the preset minimum longitudinal acceleration, maximum longitudinal acceleration, minimum lateral acceleration, and maximum lateral acceleration of the vehicle. and represent the preset maximum longitudinal and lateral speeds. represents the upper boundary of the lane, represents the lower boundary of the lane. represents the width of the vehicle.

[0094] S32. The original problem can be decomposed into two sub-problems, namely the unilateral decision-making problem of the vehicle and the update problem of the auxiliary variable .

[0095] In each iteration, the specific steps are as follows:

[0096] 1) Update the vehicle control variable : With the auxiliary variable and the dual variable fixed, independently solve the unilateral decision-making problem for each vehicle to optimize its control strategy;

[0097] 2) Update the auxiliary variable : With the vehicle control variable and the dual variable fixed, coordinate the decisions among vehicles by solving the update problem of the auxiliary variable ;

[0098] 3) Update the dual variable : According to the update rule of the Lagrange multiplier method, adjust the dual variable to gradually satisfy the constraint conditions.

[0099] It can be solved using a distributed algorithm:

[0100] (18)

[0101] (19)

[0102] (20)

[0103] (21)

[0104] In the formula, is the number of iterations of the ADMM algorithm. is the penalty parameter.

[0105] S33. The vehicle state update equation is:

[0106] (22)

[0107] (23)

[0108] (24)

[0109] (25)

[0110] In the formula, represents the longitudinal position, represents the lateral position. represents the longitudinal speed, represents the lateral speed. represents the adjacent time interval.

[0111] The trajectory planning is a dynamic process. All vehicle trajectories at time t + 1 are generated according to the state at the current time t. The generated trajectories are as Figure 4 shown.

[0112] Advantages of this embodiment:

[0113] In this embodiment, a potential function considering traffic efficiency, driving goals, comfort, and safety is established, which accurately depicts the decision-making factors of human drivers in merging and lane-changing. A stochastic potential game model is introduced to capture sensor errors. The stochastic optimization model is transformed into a deterministic model by using average approximate sampling to model the interactive game between vehicles. The established model can reflect the interactive logic of human drivers in merging and lane-changing. A two-layer programming model is used to calibrate the driving style weight factor in the model to more realistically depict human driving behavior. Finally, the stochastic potential game is combined with model predictive control and applied to the trajectory planning and control of autonomous vehicles to achieve human-like merging control. The proposed planning and control framework can improve the interaction between autonomous vehicles and human-driven vehicles in mixed traffic, thereby improving regional traffic efficiency and safety.

[0114] Embodiment 2

[0115] This embodiment also discloses a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method described in Embodiment 1 are implemented.

[0116] Embodiment 3

[0117] This embodiment also discloses a computer program product, including a computer program which, when executed by a processor, implements the steps of the method described in Embodiment 1.

[0118] The above are only the preferred specific embodiments of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in the present application should be covered by the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. A human-like merging trajectory control method for mixed traffic flow based on stochastic potential game, characterized in that It includes the following steps: Establish a stochastic potential game model considering the impact of communication delay or sensing uncertainty, determine the potential function in the stochastic potential game model, and calibrate the parameters in the stochastic potential game model using an open-source dataset; Collect the real-time state information of all vehicles within the highway merge area as the input information for trajectory planning; Based on the stochastic potential game model, construct a model predictive controller and solve it using a distributed algorithm; use the real-time vehicle state information to calculate the vehicle action combination at the next moment through the stochastic potential game model, obtain the vehicle planned trajectory, and update the vehicle state.

2. The method according to claim 1, wherein The process of establishing a stochastic potential game model considering the impact of communication delay or sensing uncertainty includes: According to the vehicle finding the best action under the stochastic potential game model, convert the original objective into a mathematical expectation with a random vector, and convert the deterministic constraint into a chance constraint; Use the sample average approximation method to convert the uncertain mathematical expectation and conditional constraint into a deterministic problem; Determine the potential function in the game model, and the potential function is a continuous and exact potential game potential function; Extract the vehicle trajectories with interactions within the highway merge area from the open-source dataset and calibrate the parameters using a model method.

3. The method according to claim 2, wherein The potential function includes self-correlation cost and coupling cost. The self-correlation cost includes passing efficiency cost, driving comfort cost, and cost of deviating from the target state. The coupling cost includes safety cost.

4. The method according to claim 2, wherein The parameter calibration uses a bilevel programming model. The upper-level model minimizes the mean absolute error between the predicted action and the real action, and the lower-level model solves the potential game equilibrium.

5. The method according to claim 1, wherein All the real-time vehicle state information includes: Collect the current lateral and longitudinal positions, lateral and longitudinal speeds, and lateral and longitudinal accelerations of the vehicle as dynamic input information; Collect the vehicle type, the lane to which the vehicle belongs, and the vehicle driving style factor as static input information; Collect the length of the merge area, the road width, and the road boundary information.

6. The method according to claim 1, wherein The steps of constructing a model predictive controller include: Establish an MPC optimization model considering vehicle dynamics constraints, road conditions, traffic rules, and the behaviors of other vehicles; Decompose the original optimization problem into a vehicle unilateral decision problem and an auxiliary variable update problem; Use a distributed algorithm for solving.

7. The method according to claim 6, wherein The steps of using a distributed algorithm for solving include: Under the condition of fixing the auxiliary variables and dual variables, independently solve the unilateral decision problem for each vehicle to optimize its control strategy; Under the condition of fixing the vehicle control variables and dual variables, coordinate the decisions among vehicles by solving the update problem of the auxiliary variables; Adjust the dual variables according to the update rule of the Lagrange multiplier method to gradually satisfy the constraint conditions.

8. The method according to claim 1, wherein The trajectory planning is a dynamic process, and all vehicle trajectories for the next moment are generated according to the state at the current moment.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-8.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1-8.

Images