Vehicle human-like lane changing control method fusing predicted trajectory and related device

Abstract

The invention belongs to the technical field of automatic driving, and discloses a vehicle human-like lane changing control method and related device.Firstly, an expert track set is constructed based on historical vehicle track data, and a candidate track set is generated by disturbing the end point state and polynomial interpolation; track features are extracted, and feature vectors are constructed in combination with environment vehicle future information; then taking the expert track as a sample, taking the candidate track as a decision space, and applying a maximum entropy principle and a feature matching mechanism to learn a humanoid lane change award function so as to reproduce expert behavior preference and select an optimal candidate track; and finally, constructing a multi-objective optimization function, optimizing trajectory parameters through piecewise nonlinear programming in a space-time dimension, generating an optimal trajectory, and outputting the optimal trajectory to a vehicle controller to execute lane changing. By adopting the method, safer, smoother, high-adaptability and high-universality human-like lane changing trajectory planning is realized, and the practical application performance of an autonomous lane changing system is remarkably improved.

Description

Technical Field

[0001] This invention belongs to the field of autonomous driving technology, and in particular relates to a vehicle-human lane-changing control method and related device that integrates predicted trajectories. Background Technology

[0002] With the rapid development of automotive intelligence technology, intelligent vehicles, leveraging advanced core technologies such as perception, trajectory planning, and tracking control, can respond to changes in the traffic environment in real time and dynamically adjust their own operating status. This demonstrates significant advantages in reducing traffic accident rates and improving road traffic efficiency, making it a research hotspot and development trend in the transportation field. Autonomous lane-changing systems, as one of the key technologies ensuring the safe and efficient operation of intelligent vehicles, function by integrating dynamic traffic environment information and completing the entire lane-changing process through safe and reasonable trajectory planning and high-precision trajectory tracking control. Among these, trajectory planning and tracking control, as core technologies of autonomous lane-changing systems, directly determine the safety, reliability, and smoothness of the lane-changing process, playing a crucial supporting role in improving the intelligence level of intelligent vehicles.

[0003] Currently, trajectory planning methods in the field of autonomous lane changing for intelligent vehicles are mainly divided into three categories: sampling and search-based methods, curve fitting-based methods, and optimization-based methods. Among them, trajectory planning methods based on sampling and search, which traverse all trajectory nodes and select the optimal trajectory according to a preset objective, have good real-time performance. Typical examples include the A* algorithm, Dijkstra's algorithm, and Rapidly-Exploring Random Tree (RRT) algorithm. However, these methods generally ignore vehicle dynamics and kinematic constraints, which can easily lead to poor continuity of the planned trajectory. Moreover, the planning efficiency decreases significantly with the increase of the number of sampling nodes, making it difficult to meet the trajectory planning requirements of actual lane-changing scenarios. Curve fitting-based methods construct a candidate trajectory set and select the best trajectory by using preset curve types (such as polynomial curves, Bezier curves, sine function curves, B-spline curves, etc.) and state sampling. They are widely used due to their simple construction and smooth trajectory. However, these methods only rely on the states of some key nodes to generate a complete trajectory, which has poor adaptability to dynamic changes in the driving environment. They usually need to be combined with optimization algorithms to complete lane-changing trajectory planning. Optimization-based trajectory planning methods construct an objective function that reflects the comprehensive performance of the trajectory and use quadratic programming, nonlinear programming, and model predictive control. Methods such as Control and MPC can solve for the optimal trajectory, but because the construction of trajectory evaluation index is highly dependent on expert experience, the obtained optimal trajectory is only effective within the expert's preset framework, which greatly limits the applicability of the method in different traffic scenarios.

[0004] It is evident that existing autonomous lane-changing trajectory planning methods for intelligent vehicles all have their own insurmountable defects, failing to simultaneously ensure trajectory safety, continuity, dynamic adaptability, and scenario universality, thus making it difficult to fully meet the practical application needs of autonomous lane-changing systems. Summary of the Invention

[0005] This invention provides a vehicle-human lane-changing control method and related device that integrates predicted trajectories. This method can take into account the safety, continuity, dynamic adaptability and scenario versatility of the trajectory, and can fully meet the practical application needs of autonomous lane-changing systems.

[0006] To achieve the above objectives, the present invention adopts the following technical solution: A vehicle-human lane-changing control method that integrates predicted trajectories includes: A set of expert trajectories for lane changing is constructed based on historical vehicle trajectory data. A set of candidate trajectories for lane changing is generated by perturbing the endpoint state of the expert trajectories and using polynomial interpolation. Trajectory features of the expert trajectories and candidate trajectories for lane changing are extracted and combined with environmental and future vehicle trajectory information to construct a trajectory feature vector. Using expert trajectory sets as real samples, candidate trajectory sets as the decision space, and trajectory feature vectors as feature inputs, a human-like lane-changing reward function is learned through the maximum entropy principle and feature matching mechanism. Based on the human-like lane-changing reward function, the candidate trajectory with the highest reward value is selected from the vehicle lane-changing candidate trajectory set as the human-like lane-changing trajectory. The human-like lane-changing reward function is used to reproduce expert behavior preferences. Based on a human-like lane-changing trajectory, a multi-objective optimization function is constructed. The trajectory parameters of the multi-objective optimization function are optimized in the spatiotemporal dimension through piecewise nonlinear programming to generate the optimal human-like lane-changing trajectory. The optimization objectives of the multi-objective optimization function include human-likeness, comfort, lane-changing efficiency, and safety. The optimal human-like lane-changing trajectory is output as the tracking trajectory to the vehicle's lateral and longitudinal controllers, so that the vehicle's lateral and longitudinal controllers can complete the human-like lane-changing control process based on the tracking trajectory.

[0007] Furthermore, the process involves constructing a vehicle lane-changing expert trajectory set based on historical vehicle trajectory data, generating a vehicle lane-changing candidate trajectory set by perturbing the endpoint state of the expert trajectories and using polynomial interpolation; extracting trajectory features from the vehicle lane-changing expert trajectory set and the vehicle lane-changing candidate trajectory set, and constructing a trajectory feature vector by combining environmental vehicle future trajectory information, including: Several sets of historical vehicle trajectory data were extracted based on the highway driving dataset; The endpoint state of the expert trajectory is perturbed and sampled, and then combined with polynomial interpolation to generate a set of candidate trajectories for vehicle lane changing, as specifically expressed below:

[0008] In the formula, Vertical position The horizontal position For lane changing time, For lane change duration, These correspond to the coefficients of each term in a fourth-degree polynomial; The coefficients of each term in the fifth-order polynomial are obtained by interpolation of the boundary states of the candidate trajectory's starting and ending points. For the candidate trajectory set of vehicle lane changing; Trajectory features are extracted from the expert trajectory set and the candidate trajectory set for lane changing, and then combined with environmental and future vehicle trajectory information to construct a trajectory feature vector. The trajectory feature vector includes trajectory dynamics features. Trajectory endpoint characteristics Interaction features with trajectory The specific expression is as follows:

[0009] In the formula, The longitudinal and lateral velocity characteristics of the trajectory; It is an acceleration characteristic; This refers to the duration of the lane change. The longitudinal velocity at the end of the trajectory; For longitudinal interaction features of the trajectory; It is a lateral interaction feature.

[0010] Furthermore, the process of learning a human-like lane-changing reward function using an expert trajectory set as the real sample, a candidate trajectory set as the decision space, and trajectory feature vectors as feature inputs, through the maximum entropy principle and feature matching mechanism; and selecting the candidate trajectory with the highest reward value from the vehicle lane-changing candidate trajectory set as the human-like lane-changing trajectory based on the human-like lane-changing reward function, includes: Based on the principle of maximum entropy, the expert trajectory set, and the maximization of the probability distribution entropy of the candidate trajectory set, the objective function is constructed, and the specific formula is as follows:

[0011] in, For candidate trajectory set The probability distribution entropy, Candidate trajectories The probability of being selected. For the first i A line of expert trajectory; Based on the constructed objective function and trajectory feature vectors, candidate trajectories are obtained by solving the Lagrange multiplier method and KKT conditions. The probability of being selected is calculated using the following formula:

[0012] in, Candidate trajectories The reward function takes the trajectory feature vector as input. ; These are the function parameters of the reward function; The natural index of trajectory reward is defined as the probability of a trajectory being selected being proportional to the natural index of trajectory reward. Let be the partition function, defined as the sum of the natural exponents of the rewards for all candidate trajectories; Objective function This is equivalent to maximizing the log-likelihood of the probability distribution of the expert trajectory set, as shown in the following formula:

[0013] Establish a nonlinear reward function for a DNN , where the objective function For the parameters of the nonlinear reward function of DNN The gradient is:

[0014] Then the model loss function LOSS for:

[0015] Based on the vehicle lane-changing expert trajectory set, the vehicle lane-changing candidate trajectory set, and the objective function, the DNN nonlinear reward function is optimized to obtain parameters. To obtain the humanoid lane-changing reward function, and select the candidate trajectory with the largest reward value as the humanoid lane-changing trajectory, as expressed below:

[0016] In the formula, and These are the longitudinal trajectory and the lateral trajectory of a human-like lane change, respectively.

[0017] Furthermore, the construction of a multi-objective optimization function based on the human-like lane-changing trajectory, and the optimization of the trajectory parameters of the multi-objective optimization function in the spatiotemporal dimension through piecewise nonlinear programming to generate the optimal human-like lane-changing trajectory, includes: The human-like lane-changing trajectory is divided into multiple trajectory segments. The human-like lane-changing trajectory is optimized through a pre-constructed segmented trajectory objective function and constraints, as specifically expressed below:

[0018] In the formula, and The first The longitudinal and transverse polynomial coefficients of each trajectory segment ;time The constraint range is , , These represent the lower and upper limits of time, respectively. A multi-objective optimization function is constructed that takes into account human-likeness, comfort, lane-changing efficiency, and safety during the vehicle lane-changing process. The specific expression is as follows:

[0019] in, and These are the objective function terms related to human-likeness, comfort, lane-changing efficiency, and safety. Weights for the objective function terms; Based on a multi-objective optimization function, lane-changing duration is introduced as an optimization parameter to optimize the lane-changing trajectory in both spatial and temporal dimensions, thereby obtaining the optimal human-like lane-changing trajectory.

[0020] Furthermore, in the process of solving the multi-objective optimization function, constraints are imposed on some trajectory states, and the specific constraint formulas are as follows: Taking the starting state of the lane change trajectory as the current state, the starting state constraints are set as follows:

[0021] in, and These are the vertical and horizontal starting positions of the candidate trajectory, respectively. Based on the human-like lane-changing trajectory, the endpoint state constraints are set as follows:

[0022] in, The duration of the humanoid lane-changing trajectory. These represent the constraint ranges for the longitudinal and lateral endpoint positions of the segmented trajectory, respectively. and The optimized trajectory for segment N is at the endpoint time. The longitudinal and lateral positions; Set the trajectory connection point state constraints as follows:

[0023] Based on the starting point state constraints, ending point state constraints, and trajectory connection point state constraints, the segmented lane-changing trajectory optimization is constructed as a nonlinear programming problem. Solving the nonlinear programming problem yields the polynomial coefficients and lane-changing duration of the segmented lane-changing trajectory, thereby obtaining the optimal human-like lane-changing trajectory.

[0024] Furthermore, the optimal human-like lane-changing trajectory is output as the tracking trajectory to the vehicle's lateral and longitudinal controllers, so that the vehicle's lateral and longitudinal controllers can complete the human-like lane-changing control process based on the tracking trajectory, including: The optimal human-like lane-changing trajectory is output as the tracking trajectory to the vehicle's lateral and longitudinal controllers. Based on the tracking trajectory, the steering angle and longitudinal acceleration of the vehicle's lateral and longitudinal controllers are obtained using a linear incremental model predictive control method and a dual PID control method, so as to realize the vehicle-like lane-changing process.

[0025] Furthermore, the step of obtaining the steering angle and longitudinal acceleration of the vehicle's lateral and longitudinal controllers based on the tracking trajectory using a linear incremental model predictive control method and a dual PID control method includes: Establish a three-degree-of-freedom vehicle dynamics model that reflects the lateral dynamic characteristics of the vehicle:

[0026] In the formula, where, For longitudinal displacement, This is a lateral displacement. For the overall vehicle quality, The yaw rate is angular velocity. , These are the longitudinal stiffnesses of the front and rear wheels, , These are the lateral stiffness of the front and rear wheels, respectively. For the front wheel steering angle, , These are the slip ratios of the front and rear wheels, respectively. For vehicles to bypass Moment of inertia of the shaft These are the wheelbases of the front and rear axles, respectively. Select system state variables as The control input is The three-degree-of-freedom vehicle dynamics model is then expressed as:

[0027] Introduce the front wheel steering angle state from the previous moment into the model state variables, and increment the front wheel steering angle. As the control input, the incremental model state and output equations are obtained as follows:

[0028] in, , , , , It is the identity matrix. To control the cycle, They are respectively At the reference point Jacobian matrix at the location, This refers to the system's bias term in discrete time. For the system's bias term in continuous time, , , , For system output variables, This is the output matrix; The system output, predicted iteratively using a three-degree-of-freedom vehicle dynamics model, is as follows:

[0029] in: , , , ; The control optimization problem for constructing a linear incremental MPC lateral trajectory tracking controller is as follows:

[0030] in, For the desired lateral trajectory, For the tracking error weight matrix, This is the corner increment weight matrix. As a relaxation factor, These are the relaxation factor weighting coefficients; The optimal incremental control sequence is obtained by repeatedly solving the control optimization problem. ; The first control quantity It is applied to vehicle control to achieve the task of tracking the lateral trajectory of vehicles in order to complete the vehicle-like lane-changing control process.

[0031] A vehicle-like lane-changing control system that integrates predicted trajectories includes: An extraction module is constructed to build a set of expert trajectories for lane changing based on historical vehicle trajectory data. A set of candidate trajectories for lane changing is generated by perturbing the endpoint state of the expert trajectories and using polynomial interpolation. Trajectory features of the set of expert trajectories for lane changing and the set of candidate trajectories for lane changing are extracted and combined with environmental vehicle future trajectory information to construct a trajectory feature vector. The learning module is used to learn a human-like lane-changing reward function by using the expert trajectory set as the real sample, the candidate trajectory set as the decision space, and the trajectory feature vector as the feature input, through the maximum entropy principle and feature matching mechanism. Based on the human-like lane-changing reward function, the candidate trajectory with the highest reward value is selected from the vehicle lane-changing candidate trajectory set as the human-like lane-changing trajectory. The human-like lane-changing reward function is used to reproduce expert behavior preferences. The optimization module is used to construct a multi-objective optimization function based on a human-like lane-changing trajectory. It optimizes the trajectory parameters of the multi-objective optimization function in the spatiotemporal dimension through piecewise nonlinear programming to generate the optimal human-like lane-changing trajectory. The optimization objectives of the multi-objective optimization function include human-likeness, comfort, lane-changing efficiency, and safety. The control module is used to output the optimal human-like lane-changing trajectory as the tracking trajectory to the vehicle's lateral and longitudinal controllers, so that the vehicle's lateral and longitudinal controllers can complete the human-like lane-changing control process based on the tracking trajectory.

[0032] A vehicle-like lane-changing control device that integrates predicted trajectories includes: Memory, used to store computer programs; A processor is used to implement the steps of the above-described vehicle-human lane-changing control method for fusing predicted trajectories when executing the computer program.

[0033] A computer-readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of the above-described vehicle-human lane-changing control method for fusing predicted trajectories.

[0034] Compared with the prior art, the present invention has the following beneficial effects: This invention provides a human-like lane-changing control method for vehicles that integrates predicted trajectories. First, an expert trajectory set is constructed based on historical vehicle trajectory data. A candidate trajectory set is generated by perturbing the endpoint state and using polynomial interpolation. Trajectory features are extracted and combined with environmental and future vehicle information to construct feature vectors. Then, using expert trajectories as samples and candidate trajectories as the decision space, the maximum entropy principle and feature matching mechanism are applied to learn a human-like lane-changing reward function to reproduce expert behavioral preferences and select the optimal candidate trajectory. Finally, a multi-objective optimization function is constructed, and the trajectory parameters are optimized in the spatiotemporal dimension through piecewise nonlinear programming to generate the optimal trajectory, which is then output to the vehicle controller for lane changing. This method uses expert data to drive reward learning, simulating human driving preferences to ensure human-like trajectory performance. The feature matching mechanism integrates dynamic environmental information to enhance adaptability. Multi-objective optimization comprehensively considers human-likeness, comfort, lane-changing efficiency, and safety, improving trajectory continuity and robustness. This method effectively solves the problems of existing technologies, such as neglecting vehicle constraints, poor dynamic adaptability, and reliance on expert experience. It achieves safer, smoother, more adaptable, and more universal human-like lane-changing trajectory planning, significantly improving the practical application performance of autonomous lane-changing systems. Attached Figure Description

[0035] Figure 1 A flowchart illustrating the implementation of a vehicle-human lane-changing control method based on fused predicted trajectories, provided in an embodiment of the present invention; Figure 2 This invention provides a vehicle-layered, human-like lane-changing strategy in its embodiments. Figure 3 Expert trajectory set diagram provided for embodiments of the present invention; Figure 4 This is a block diagram of the lane-changing trajectory planning model provided in an embodiment of the present invention; Figure 5 A structural diagram of a longitudinal trajectory tracking controller provided in an embodiment of the present invention; Figure 6 A flowchart of a vehicle-human lane-changing control method based on fused predicted trajectories provided in an embodiment of the present invention; Figure 7 This is a schematic diagram of a vehicle-like lane-changing control system that integrates predicted trajectories, provided as an embodiment of the present invention. Detailed Implementation

[0036] To further understand the content of this invention, the invention will be described in detail below with reference to the accompanying drawings and specific embodiments. It should be understood that the embodiments are merely illustrative and not limiting of the invention.

[0037] The following explains the technical terms used in this invention: KKT conditions: The full name is Karush-Kuhn-Tucker conditions, which are important tools in optimization theory for solving constrained optimization problems.

[0038] Lagrange multiplier method: The Lagrange multiplier method is a mathematical method for finding extrema in equality-constrained optimization problems.

[0039] IRL stands for Inverse Reinforcement Learning, a machine learning method that infers the implicit reward function by observing the behavior of experts.

[0040] HighD dataset: A high-precision dataset designed specifically for autonomous driving research in highway scenarios.

[0041] MPC stands for Model Predictive Control, which is an optimization control algorithm based on a system model.

[0042] PID: A closed-loop control algorithm based on error feedback. It dynamically adjusts the control input by combining proportional, integral and derivative components, so that the output of the controlled object can track the target value quickly, accurately and stably.

[0043] This embodiment provides a vehicle-like lane-changing control method that integrates predicted trajectories. This method enables intelligent vehicles to change lanes safely and stably in dynamic traffic scenarios on highways, while ensuring that lane-changing decisions and trajectories conform to the driving habits of real drivers.

[0044] like Figure 6 As shown in the figure, this embodiment provides a vehicle-human lane-changing control method that integrates predicted trajectories. The specific steps are as follows: A set of expert trajectories for lane changing is constructed based on historical vehicle trajectory data. A set of candidate trajectories for lane changing is generated by perturbing the endpoint state of the expert trajectories and using polynomial interpolation. Trajectory features of the expert trajectories and candidate trajectories for lane changing are extracted and combined with environmental and future vehicle trajectory information to construct a trajectory feature vector. Using expert trajectory sets as real samples, candidate trajectory sets as the decision space, and trajectory feature vectors as feature inputs, a human-like lane-changing reward function is learned through the maximum entropy principle and feature matching mechanism. Based on the human-like lane-changing reward function, the candidate trajectory with the highest reward value is selected from the vehicle lane-changing candidate trajectory set as the human-like lane-changing trajectory. The human-like lane-changing reward function is used to reproduce expert behavior preferences. Based on a human-like lane-changing trajectory, a multi-objective optimization function is constructed. The trajectory parameters of the multi-objective optimization function are optimized in the spatiotemporal dimension through piecewise nonlinear programming to generate the optimal human-like lane-changing trajectory. The optimization objectives of the multi-objective optimization function include human-likeness, comfort, lane-changing efficiency, and safety. The optimal human-like lane-changing trajectory is output as the tracking trajectory to the vehicle's lateral and longitudinal controllers, so that the vehicle's lateral and longitudinal controllers can complete the human-like lane-changing control process based on the tracking trajectory.

[0045] The lane-changing method provided in this embodiment will be further explained below with reference to the accompanying drawings: Combination Figure 1 and Figure 2 As shown, this embodiment provides a vehicle-human lane-changing control method that integrates predicted trajectories, including: S1. Establish a vehicle lane-changing expert trajectory set based on the HighD dataset; then, while keeping the starting point of the expert trajectory consistent with the environmental conditions, generate a vehicle lane-changing candidate trajectory set by perturbation sampling and polynomial interpolation of the endpoint state of the expert trajectory; finally, extract features and uniformly represent the vehicle lane-changing expert trajectory set and its corresponding vehicle lane-changing candidate trajectory set, and construct trajectory feature vectors by combining environmental vehicle future trajectory information. S2. The maximum entropy deep IRL model uses the expert trajectory set as the real behavior sample, constructs the optional decision space with the candidate trajectory set, and realizes the unified feature representation with the trajectory feature vector. Based on the maximum entropy principle and feature matching mechanism, it learns a human-like lane-changing reward function that can reproduce the expert's behavioral preferences, and selects the trajectory with the highest reward value from the candidate trajectory set based on the reward function to realize the human-like lane-changing trajectory. S3. Using the human-like lane-changing trajectory as a reference trajectory, a multi-objective optimization function is constructed that simultaneously considers human-likeness, comfort, lane-changing efficiency and safety. A piecewise nonlinear programming method is then used to optimize the trajectory parameters in the spatial and temporal dimensions, thereby obtaining the optimal human-like lane-changing trajectory that balances multiple indicators. S4. The optimal human-like lane-changing trajectory provides a reference trajectory for the control layer. The vehicle's lateral and longitudinal controllers, based on the linear incremental model predictive control method and the dual PID control method, use this trajectory as a reference. Through the coordinated control of steering angle and longitudinal acceleration, the vehicle can accurately track the planned trajectory, thereby achieving a safe and smooth human-like lane-changing process.

[0046] In this embodiment, the specific steps of S1 are as follows: Step 1: Based on the HighD dataset, a total of 447 sets of vehicle lane-changing trajectory data were extracted, including the vehicle's own lane-changing trajectory and the trajectory of surrounding vehicles. To ensure the continuity of the vehicle's lane-changing trajectory, a fifth-order polynomial interpolation was used to construct a vehicle lane-changing expert trajectory set, such as... Figure 3 Show.

[0047] Step 2: For the candidate trajectory set, assume the expert trajectory Corresponding candidate trajectory set Total If there are 10 candidate trajectories, then the candidate trajectory set is... It can be represented as:

[0048] in, Indicates the first i The first of the expert trajectories j Candidate trajectories.

[0049] Step 3: To ensure trajectory continuity and comfort, candidate trajectories are constructed based on fourth and fifth-order polynomials in the Frenet coordinate system. As shown in the following formula:

[0050] in, Vertical position The horizontal position For lane changing time, For lane change duration, and The coefficients are divided into fourth and fifth degree polynomials, which can be solved by interpolation of the boundary states of the candidate trajectory's starting and ending points.

[0051] Step 4: Establish trajectory feature vectors that take into account the future trajectory of vehicles in the environment, so as to enhance the inverse reinforcement learning model's understanding of the driving environment and improve lane changing safety.

[0052] candidate trajectories For example, this embodiment divides trajectory features into trajectory dynamic features, trajectory endpoint features, and trajectory interaction features: (1) Trajectory dynamics characteristics Includes longitudinal and lateral velocity characteristics of the trajectory. and acceleration characteristics :

[0053] (2) Trajectory endpoint characteristics Includes lane change duration longitudinal velocity at the end of the trajectory :

[0054] (3) Trajectory interaction features The pre-established IPF-GCN model is used to predict the future trajectory of vehicles in the environment. The specific formula is as follows:

[0055] in, For the first k Predicted trajectories of environmental vehicles. and The first k The longitudinal and lateral predicted trajectories of a vehicle in the environment. The number of vehicles used in the environment.

[0056] Based on the longitudinal and lateral distances between the vehicle lane-changing trajectory and the predicted trajectory of the vehicle in the environment, trajectory interaction features are constructed, and the specific formula is as follows:

[0057] in, and These are respectively the longitudinal interaction features and the lateral interaction features of the trajectory. To predict the length of the time domain, This represents the total length of the time step.

[0058] Ultimately, the trajectory feature vector as follows:

[0059] For example, such as Figure 4 As shown, the specific steps of S2 are as follows: Step 1: Using the candidate trajectory set For example, based on the principle of maximum entropy, maximizing the probability distribution entropy of the candidate trajectory set can be constructed as the following optimization problem:

[0060] in, For candidate trajectory set The probability distribution entropy, Candidate trajectories The probability of being selected. For the first i A path for experts.

[0061] Step 2: Using the Lagrange multiplier method and KKT conditions, candidate trajectories are obtained. The probability of being selected is as follows:

[0062] in, Candidate trajectories The reward function takes the trajectory feature vector as its input. ; These are the function parameters of the reward function; The natural index of the trajectory reward is proportional to the probability of a trajectory being selected. Let be the partition function, which is defined as the sum of the natural reward exponents of all candidate trajectories.

[0063] Step 3: To facilitate solution and analysis, the objective function of maximizing the entropy of the probability distribution of the candidate trajectory set is equivalently transformed into maximizing the log-likelihood of the probability distribution of the expert trajectory set, as shown in the following equation:

[0064] Step 4: Establish the DNN nonlinear reward function This is to improve the model's learning ability. Objective function For the parameters of the nonlinear reward function of DNN The gradient is:

[0065] That is, the model loss function is:

[0066] Step 5: Based on the constructed expert trajectory set and candidate trajectory set, optimize the parameters of the DNN nonlinear reward function by maximizing the objective function represented by the probability distribution entropy of the candidate trajectory set. The humanoid lane-changing reward function is obtained, and the candidate trajectory with the largest reward value is selected as the humanoid lane-changing trajectory. ,in and These are the longitudinal trajectory and the lateral trajectory of a human-like lane change, respectively.

[0067] For example, in this embodiment, the specific steps of S3 are as follows: Step 1: Divide the humanoid lane-changing trajectory into... The number of trajectory segments is calculated by designing segmented trajectory objective functions and constraints to optimize the lane-changing trajectory. Depending on computing power and lane-changing time, the segmented trajectory is as follows:

[0068] in, and The first The longitudinal and transverse polynomial coefficients of each trajectory segment ;time The constraint range is , , These represent the lower and upper limits of time, respectively.

[0069] Step 2: To ensure that the lane-changing trajectory planning of intelligent vehicles conforms to human driving habits and can adapt to dynamic changes in the traffic environment, guaranteeing lane-changing comfort, efficiency, and safety, the trajectory optimization objective function is designed as follows:

[0070] in, and These are the objective function terms related to human-likeness, comfort, lane-changing efficiency, and safety. The weights for the objective function terms are as follows: (1) Human-like nature: Changing the trajectory of human-like beings As a baseline trajectory, the distance deviation between the segmented trajectory and the humanoid lane-changing trajectory is used to measure humanoid characteristics. The specific formula is as follows:

[0071] in, and The first The longitudinal and lateral lane-changing trajectories of each trajectory segment; (2) Comfort: Excessive acceleration and frequent acceleration and deceleration during lane changes can cause discomfort to passengers. Therefore, comfort is measured by the acceleration and jerk of the segmented trajectory. The specific formula is as follows:

[0072] (3) Lane changing efficiency: Lane changing efficiency is measured by lane changing time, and the specific formula is as follows:

[0073] (4) Safety: Time to Collision (TC) is an important indicator for measuring vehicle driving safety. Therefore, a predicted TTC considering the vehicle's future trajectory was established. The predicted TTC is calculated by combining the planned trajectory of the vehicle with the predicted trajectories of the vehicles in front and behind. The deviation between the actual vehicle speed and the target vehicle speed is set as the objective function term for measuring lane-changing safety, enabling early perception of collision risks and further improving vehicle driving safety. The specific formula is as follows:

[0074] in, These are the predicted collision times related to the predicted trajectories of the preceding and following vehicles; and the predicted collision times of the current vehicle. When the speed decreases, it reduces the target speed of the vehicle behind, improving lane-changing safety; when the following vehicle anticipates... When the speed increases, it increases the target speed of the vehicle, reduces the risk of collision from vehicles behind, and improves lane traffic efficiency. For lane reference speed, These represent the speeds of the vehicle itself, the vehicle in front, and the vehicle behind.

[0075] Step 3: Lane-changing trajectories possess both temporal and spatial characteristics. Using polynomial coefficients as optimization parameters can only adjust the spatial distribution of the trajectory, but not its temporal distribution. Therefore, this embodiment adds the lane-changing duration as an optimization parameter to optimize the lane-changing trajectory from both spatial and temporal dimensions. When solving the optimization problem, to ensure the continuity and feasibility of the segmented trajectories, it is necessary to constrain the state of some trajectories. The specific formula is as follows: (1) The starting state of the lane-changing trajectory is the current state, and the starting state constraints are set as follows:

[0076] in, and These represent the longitudinal and lateral starting positions of the candidate trajectory, respectively.

[0077] (2) To ensure the human-like nature of the segmented trajectory, the endpoint state constraints are set based on the human-like lane-changing trajectory as follows:

[0078] in, The duration of the humanoid lane-changing trajectory. These represent the constraint ranges for the longitudinal and lateral endpoint positions of the segmented trajectory, respectively. and The optimized trajectory for segment N is at the endpoint time. The longitudinal and lateral positions.

[0079] (3) To ensure the continuity of the segmented trajectory, the following state constraints are set for the trajectory connection points:

[0080] Step 4: Based on the above analysis, this embodiment constructs the segmented lane-changing trajectory optimization as a nonlinear programming problem. By solving the following optimization problem, the polynomial coefficients and lane-changing duration of the segmented lane-changing trajectory are obtained, thus realizing human-like optimal lane-changing trajectory planning.

[0081]

[0082] For example, in this embodiment, the specific steps of S4 are as follows: Step 1: Establish a three-degree-of-freedom vehicle dynamics model that accurately reflects the lateral dynamic characteristics of the vehicle:

[0083] in, For longitudinal displacement, This is a lateral displacement. For the overall vehicle quality, The yaw rate is angular velocity. , These are the longitudinal stiffnesses of the front and rear wheels, , These are the lateral stiffness of the front and rear wheels, respectively. For the front wheel steering angle, , These are the slip ratios of the front and rear wheels, respectively. For vehicles to bypass Moment of inertia of the shaft These are the wheelbases of the front and rear axles, respectively.

[0084] Select system state variables as The control input is The vehicle dynamics model can be represented as:

[0085] Step 2: Considering that using the front wheel steering angle as the control input cannot constrain the change in steering angle increment, leading to abrupt changes in the front wheel steering angle and affecting the comfort of the vehicle during lane changes, the front wheel steering angle state of the previous moment is included in the model's state variables, and the front wheel steering angle increment is used as the control input. As the control input, the following incremental model state and output equations are derived:

[0086] in, , , , , It is the identity matrix. To control the cycle, They are respectively At the reference point Jacobian matrix at the location, This refers to the system's bias term in discrete time. For the system's bias term in continuous time, , , , For system output variables, This is the output matrix.

[0087] When the prediction time domain of the model prediction controller is Control time domain is To simplify calculations, we assume the following:

[0088] The system output can be obtained through model iteration prediction:

[0089] in: , , , .

[0090] Therefore, the linear incremental MPC lateral trajectory tracking controller is constructed as follows:

[0091] in, For the desired lateral trajectory, For the tracking error weight matrix, This is the corner increment weight matrix. As a relaxation factor, The relaxation factor is the weighting coefficient. The relaxation factor can ensure that the model can find a solution in each solution process.

[0092] During lane change trajectory tracking, the optimal incremental control sequence is obtained by repeatedly solving the above optimization problem. and the first control quantity It is applied to vehicle control to achieve the task of tracking the lateral trajectory of the vehicle.

[0093] Step 3: Establish a longitudinal trajectory tracking controller based on dual PID control, the block diagram of which is as follows. Figure 5 As shown, the upper-level PID position controller takes the deviation between the vehicle's actual position and the planned trajectory as input, and outputs the speed deviation and compensation amount; then, the lower-level PID speed controller combines the speed deviation and compensation amount to calculate the target acceleration and apply it to the vehicle system.

[0094] like Figure 7As shown, this embodiment also provides a vehicle-like lane-changing control system that integrates predicted trajectories, including: a construction and extraction module, used to construct a vehicle lane-changing expert trajectory set based on historical vehicle trajectory data, and generate a vehicle lane-changing candidate trajectory set by perturbing the endpoint state of the expert trajectories and using polynomial interpolation; extracting trajectory features from the vehicle lane-changing expert trajectory set and the vehicle lane-changing candidate trajectory set, and constructing a trajectory feature vector by combining environmental vehicle future trajectory information; a learning module, used to learn a human-like lane-changing reward function by using the expert trajectory set as real samples, the candidate trajectory set as the decision space, and the trajectory feature vector as feature input, through the maximum entropy principle and feature matching mechanism; and based on the human-like lane-changing reward function... The excitation function selects the candidate trajectory with the highest reward value from the set of candidate trajectories for vehicle lane changing as the human-like lane changing trajectory; wherein, the human-like lane changing reward function is used to reproduce expert behavioral preferences; the optimization module is used to construct a multi-objective optimization function based on the human-like lane changing trajectory, and optimize the trajectory parameters of the multi-objective optimization function in the spatiotemporal dimension through piecewise nonlinear programming to generate the optimal human-like lane changing trajectory; wherein, the optimization objectives of the multi-objective optimization function include human-likeness, comfort, lane changing efficiency and safety; the control module is used to output the optimal human-like lane changing trajectory as the tracking trajectory to the vehicle's lateral and longitudinal controllers, so that the vehicle's lateral and longitudinal controllers can complete the human-like lane changing control process of the vehicle according to the tracking trajectory.

[0095] The present invention also provides a vehicle-like lane-changing control device that integrates predicted trajectories, comprising: a memory for storing a computer program; and a processor for executing the computer program to implement the steps of the vehicle-like lane-changing control method that integrates predicted trajectories.

[0096] The present invention also provides a computer program product, including a computer program / instructions, which, when executed by a processor, implement the steps of the vehicle-human lane-changing control method for fusing predicted trajectories.

[0097] When the processor executes the computer program, it implements the above-mentioned steps for vehicle-like lane-changing control based on fused predicted trajectories. For example: constructing a vehicle lane-changing expert trajectory set based on historical vehicle trajectory data; generating a vehicle lane-changing candidate trajectory set by perturbing the endpoint states of the expert trajectories and using polynomial interpolation; extracting trajectory features from the vehicle lane-changing expert trajectory set and the vehicle lane-changing candidate trajectory set, and constructing a trajectory feature vector by combining environmental vehicle future trajectory information; using the expert trajectory set as the real sample, the candidate trajectory set as the decision space, and the trajectory feature vector as the feature input, learning the human-like lane-changing reward function through the maximum entropy principle and feature matching mechanism; and based on the human-like... A lane-changing reward function selects the candidate trajectory with the highest reward value from the set of candidate lane-changing trajectories as the human-like lane-changing trajectory; wherein, the human-like lane-changing reward function is used to reproduce expert behavioral preferences; a multi-objective optimization function is constructed based on the human-like lane-changing trajectory, and the trajectory parameters of the multi-objective optimization function are optimized in the spatiotemporal dimension through piecewise nonlinear programming to generate the optimal human-like lane-changing trajectory; wherein, the optimization objectives of the multi-objective optimization function include human-likeness, comfort, lane-changing efficiency, and safety; the optimal human-like lane-changing trajectory is output as the tracking trajectory to the vehicle's lateral and longitudinal controllers, so that the vehicle's lateral and longitudinal controllers can complete the human-like lane-changing process of the vehicle according to the tracking trajectory.

[0098] For example, the computer program can be divided into one or more modules / units, which are stored in the memory and executed by the processor to complete the present invention. The one or more modules / units can be a series of computer program instruction segments capable of performing preset functions, the instruction segments describing the execution process of the computer program in the vehicle-like lane-changing control device for fusion predicted trajectories. For example, the computer program can be divided into a construction and extraction module, a learning module, an optimization module, and a control module, with the following specific functions: The construction and extraction module is used to construct a vehicle lane-changing expert trajectory set based on historical vehicle trajectory data, and generate a vehicle lane-changing candidate trajectory set by perturbing the endpoint state of the expert trajectories and using polynomial interpolation; extracting trajectory features from the vehicle lane-changing expert trajectory set and the vehicle lane-changing candidate trajectory set, and constructing a trajectory feature vector by combining environmental vehicle future trajectory information; the learning module is used to learn the human-like lane-changing reward function using the expert trajectory set as the real sample, the candidate trajectory set as the decision space, and the trajectory feature vector as the feature input, through the maximum entropy principle and feature matching mechanism; based on... A human-like lane-changing reward function selects the candidate trajectory with the highest reward value from the set of candidate lane-changing trajectories as the human-like lane-changing trajectory; wherein, the human-like lane-changing reward function is used to reproduce expert behavioral preferences; an optimization module is used to construct a multi-objective optimization function based on the human-like lane-changing trajectory, and optimize the trajectory parameters of the multi-objective optimization function in the spatiotemporal dimension through piecewise nonlinear programming to generate the optimal human-like lane-changing trajectory; wherein, the optimization objectives of the multi-objective optimization function include human-likeness, comfort, lane-changing efficiency, and safety; a control module is used to output the optimal human-like lane-changing trajectory as the tracking trajectory to the vehicle's lateral and longitudinal controllers, so that the vehicle's lateral and longitudinal controllers can complete the human-like lane-changing control process of the vehicle according to the tracking trajectory.

[0099] The vehicle-to-human lane-changing control device that integrates predicted trajectories can be a computing device such as a desktop computer, laptop, handheld computer, or cloud server. This device may include, but is not limited to, processors and memory. Those skilled in the art will understand that the above are examples of vehicle-to-human lane-changing control devices that integrate predicted trajectories and do not constitute a limitation on such devices. They may include more components than described above, or combine certain components, or use different components. For example, the vehicle-to-human lane-changing control device that integrates predicted trajectories may also include input / output devices, network access devices, buses, etc.

[0100] The processor referred to can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor can be a microprocessor, or any conventional processor. The processor is the control center of the vehicle-human lane-changing control system for the fused predicted trajectory, connecting various parts of the entire fused predicted trajectory vehicle-human lane-changing control device via various interfaces and lines.

[0101] The memory can be used to store the computer program and / or modules. The processor implements various functions of the vehicle-human lane-changing control device with fused predicted trajectory by running or executing the computer program and / or modules stored in the memory and calling the data stored in the memory.

[0102] The memory may primarily include a program storage area and a data storage area. The program storage area may store the operating system and at least one application program required for a function (such as sound playback, image playback, etc.). The data storage area may store data created based on the use of the mobile phone (such as audio data, phonebook, etc.). Furthermore, the memory may include high-speed random access memory and non-volatile memory, such as hard disks, RAM, plug-in hard disks, smart media cards (SMC), secure digital cards (SD cards), flash cards, at least one disk storage device, flash memory device, or other volatile solid-state storage devices.

[0103] The present invention also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the vehicle-human lane-changing control method based on fused predicted trajectories.

[0104] If the integrated module / unit of the vehicle-human lane-changing control system for the fusion prediction trajectory is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium.

[0105] Based on this understanding, all or part of the processes in the vehicle-human lane-changing control method for fusing predicted trajectories described above can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the vehicle-human lane-changing control method for fusing predicted trajectories described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or a preset intermediate form, etc.

[0106] The computer-readable storage medium may include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium, etc.

[0107] It should be noted that the content contained in the computer-readable storage medium may be appropriately increased or decreased according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable storage medium does not include electrical carrier signals and telecommunication signals.

[0108] In summary, this embodiment provides a vehicle-pedestrian lane-changing control method that integrates predicted trajectories, which has the following advantages compared to traditional control methods: First, by incorporating vehicle trajectory prediction into the establishment of candidate trajectory sets, trajectory feature vectors, and optimization objective functions, this invention not only significantly improves the training efficiency and accuracy of the maximum entropy deep inverse reinforcement learning (IRL) model, but also enhances the model's adaptation to and understanding of dynamic driving environment changes, thereby effectively improving the safety of vehicle lane changing processes.

[0109] Secondly, in the trajectory tracking control layer, a lateral trajectory tracking controller based on linear incremental MPC and a longitudinal trajectory tracking controller based on dual PID were constructed respectively, which improved the vehicle's tracking accuracy and stability of the planned path.

[0110] The above embodiments are merely one of the implementation methods for achieving the technical solution of the present invention. The scope of protection claimed by the present invention is not limited to this embodiment, but also includes any variations, substitutions and other implementation methods that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention.

[0111] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, those skilled in the art should understand that modifications or equivalent substitutions can still be made to the specific implementation of the present invention. Any modifications or equivalent substitutions that do not depart from the spirit and scope of the present invention should be covered within the protection scope of the present invention.

Claims

1. A vehicle-like lane-changing control method that integrates predicted trajectories, characterized in that, include: A set of expert trajectories for lane changing is constructed based on historical vehicle trajectory data. A set of candidate trajectories for lane changing is generated by perturbing the endpoint state of the expert trajectories and using polynomial interpolation. Extract trajectory features from the expert trajectory set and the candidate trajectory set for lane changing of vehicles, and construct trajectory feature vectors by combining environmental and future vehicle trajectory information; Using expert trajectory sets as real samples, candidate trajectory sets as decision space, and trajectory feature vectors as feature inputs, a human-like lane-changing reward function is learned through the maximum entropy principle and feature matching mechanism. Based on the human-like lane-changing reward function, the candidate trajectory with the highest reward value is selected from the vehicle lane-changing candidate trajectory set as the human-like lane-changing trajectory; wherein, the human-like lane-changing reward function is used to reproduce expert behavioral preferences; Based on a human-like lane-changing trajectory, a multi-objective optimization function is constructed. The trajectory parameters of the multi-objective optimization function are optimized in the spatiotemporal dimension through piecewise nonlinear programming to generate the optimal human-like lane-changing trajectory. The optimization objectives of the multi-objective optimization function include human-likeness, comfort, lane-changing efficiency, and safety. The optimal human-like lane-changing trajectory is output as the tracking trajectory to the vehicle's lateral and longitudinal controllers, so that the vehicle's lateral and longitudinal controllers can complete the human-like lane-changing control process based on the tracking trajectory.

2. The vehicle-human lane-changing control method based on fused predicted trajectories according to claim 1, characterized in that, The process involves constructing a set of expert trajectories for lane changing based on historical vehicle trajectory data, and generating a set of candidate trajectories for lane changing by perturbing the endpoint state of the expert trajectories and using polynomial interpolation. Trajectory features are extracted from the expert trajectory set and the candidate trajectory set for lane changing, and then combined with environmental and future vehicle trajectory information to construct a trajectory feature vector, including: Several sets of historical vehicle trajectory data were extracted based on the highway driving dataset; The endpoint state of the expert trajectory is perturbed and sampled, and then combined with polynomial interpolation to generate a set of candidate trajectories for vehicle lane changing, as specifically expressed below: In the formula, Vertical position The horizontal position For lane changing time, For lane change duration, These correspond to the coefficients of each term in a fourth-degree polynomial; The coefficients of each term in the fifth-order polynomial are obtained by interpolation of the boundary states of the candidate trajectory's starting and ending points. For the candidate trajectory set of vehicle lane changing; Trajectory features are extracted from the expert trajectory set and the candidate trajectory set for lane changing, and then combined with environmental and future vehicle trajectory information to construct a trajectory feature vector. The trajectory feature vector includes trajectory dynamics features. Trajectory endpoint characteristics Interaction features with trajectory The specific expression is as follows: In the formula, The longitudinal and lateral velocity characteristics of the trajectory; It is an acceleration characteristic; This refers to the duration of the lane change. The longitudinal velocity at the end of the trajectory; For longitudinal interaction features of the trajectory; It is a lateral interaction feature.

3. The vehicle-human lane-changing control method based on fused predicted trajectories according to claim 2, characterized in that, The method uses expert trajectory sets as real samples, candidate trajectory sets as decision space, and trajectory feature vectors as feature inputs to learn a human-like lane-changing reward function through the maximum entropy principle and feature matching mechanism. Based on the human-like lane-changing reward function, the candidate trajectory with the highest reward value is selected from the vehicle lane-changing candidate trajectory set as the human-like lane-changing trajectory, including: Based on the principle of maximum entropy, the expert trajectory set, and the maximization of the probability distribution entropy of the candidate trajectory set, the objective function is constructed, and the specific formula is as follows: in, For candidate trajectory set The probability distribution entropy, Candidate trajectories The probability of being selected. For the first i A line of expert trajectory; Based on the constructed objective function and trajectory feature vectors, candidate trajectories are obtained by solving the Lagrange multiplier method and KKT conditions. The probability of being selected is calculated using the following formula: in, Candidate trajectories The reward function takes the trajectory feature vector as input. ; These are the function parameters of the reward function; The natural index of trajectory reward is defined as the probability of a trajectory being selected being proportional to the natural index of trajectory reward. Let be the partition function, defined as the sum of the natural exponents of the rewards for all candidate trajectories; Objective function This is equivalent to maximizing the log-likelihood of the probability distribution of the expert trajectory set, as shown in the following formula: Establish a nonlinear reward function for a DNN , where the objective function For the parameters of the nonlinear reward function of DNN The gradient is: Then the model loss function LOSS for: Based on the vehicle lane-changing expert trajectory set, the vehicle lane-changing candidate trajectory set, and the objective function, the DNN nonlinear reward function is optimized to obtain parameters. To obtain the humanoid lane-changing reward function, and select the candidate trajectory with the largest reward value as the humanoid lane-changing trajectory, as expressed below: In the formula, and These are the longitudinal trajectory and the lateral trajectory of a human-like lane change, respectively.

4. The vehicle-human lane-changing control method based on fused predicted trajectories according to claim 3, characterized in that, The process involves constructing a multi-objective optimization function based on a human-like lane-changing trajectory, and optimizing the trajectory parameters of the multi-objective optimization function in the spatiotemporal dimension through piecewise nonlinear programming to generate the optimal human-like lane-changing trajectory, including: The human-like lane-changing trajectory is divided into multiple trajectory segments. The human-like lane-changing trajectory is optimized through a pre-constructed segmented trajectory objective function and constraints, as specifically expressed below: In the formula, and The first The longitudinal and transverse polynomial coefficients of each trajectory segment ;time The constraint range is , , These represent the lower and upper limits of time, respectively. A multi-objective optimization function is constructed that takes into account human-likeness, comfort, lane-changing efficiency, and safety during the vehicle lane-changing process. The specific expression is as follows: in, and These are the objective function terms related to human-likeness, comfort, lane-changing efficiency, and safety. Weights for the objective function terms; Based on a multi-objective optimization function, lane-changing duration is introduced as an optimization parameter to optimize the lane-changing trajectory in both spatial and temporal dimensions, thereby obtaining the optimal human-like lane-changing trajectory.

5. The vehicle-human lane-changing control method based on fused predicted trajectories according to claim 4, characterized in that, In solving the multi-objective optimization function, constraints are imposed on some trajectory states. The specific constraint formulas are as follows: Taking the starting state of the lane change trajectory as the current state, the starting state constraints are set as follows: in, and These are the vertical and horizontal starting positions of the candidate trajectory, respectively. Based on the human-like lane-changing trajectory, the endpoint state constraints are set as follows: in, The duration of the humanoid lane-changing trajectory. These represent the constraint ranges for the longitudinal and lateral endpoint positions of the segmented trajectory, respectively. and The optimized trajectory for segment N is at the endpoint time. The longitudinal and lateral positions; Set the trajectory connection point state constraints as follows: Based on the starting point state constraints, ending point state constraints, and trajectory connection point state constraints, the segmented lane-changing trajectory optimization is constructed as a nonlinear programming problem. Solving the nonlinear programming problem yields the polynomial coefficients and lane-changing duration of the segmented lane-changing trajectory, thereby obtaining the optimal human-like lane-changing trajectory.

6. The vehicle-human lane-changing control method based on fused predicted trajectories according to claim 1, characterized in that, The optimal human-like lane-changing trajectory is output as the tracking trajectory to the vehicle's lateral and longitudinal controllers, so that the vehicle's lateral and longitudinal controllers can complete the human-like lane-changing control process based on the tracking trajectory, including: The optimal human-like lane-changing trajectory is output as the tracking trajectory to the vehicle's lateral and longitudinal controllers. Based on the tracking trajectory, the steering angle and longitudinal acceleration of the vehicle's lateral and longitudinal controllers are obtained using a linear incremental model predictive control method and a dual PID control method, so as to realize the vehicle-like lane-changing process.

7. The vehicle-human lane-changing control method based on fused predicted trajectories according to claim 1, characterized in that, The process of obtaining the steering angle and longitudinal acceleration of the vehicle's lateral and longitudinal controllers based on the tracking trajectory using a linear incremental model predictive control method and a dual PID control method includes: Establish a three-degree-of-freedom vehicle dynamics model that reflects the lateral dynamic characteristics of the vehicle: In the formula, where, For longitudinal displacement, This is a lateral displacement. For the overall vehicle quality, The yaw rate is angular velocity. , These are the longitudinal stiffnesses of the front and rear wheels, , These are the lateral stiffness of the front and rear wheels, respectively. For the front wheel steering angle, , These are the slip ratios of the front and rear wheels, respectively. For vehicles to bypass Moment of inertia of the shaft These are the wheelbases of the front and rear axles, respectively. Select system state variables as The control input is The three-degree-of-freedom vehicle dynamics model is then expressed as: Introduce the front wheel steering angle state from the previous moment into the model state variables, and increment the front wheel steering angle. As the control input, the incremental model state and output equations are obtained as follows: in, , , , , It is the identity matrix. To control the cycle, They are respectively At the reference point Jacobian matrix at the location, This refers to the system's bias term in discrete time. For the system's bias term in continuous time, , , , For system output variables, This is the output matrix; The system output, predicted iteratively using a three-degree-of-freedom vehicle dynamics model, is as follows: in: ， ， ， ； The control optimization problem for constructing a linear incremental MPC lateral trajectory tracking controller is as follows: in, For the desired lateral trajectory, For the tracking error weight matrix, This is the corner increment weight matrix. As a relaxation factor, These are the relaxation factor weighting coefficients; The optimal incremental control sequence is obtained by repeatedly solving the control optimization problem. ； The first control quantity It is applied to vehicle control to achieve the task of tracking the lateral trajectory of vehicles in order to complete the vehicle-like lane-changing control process.

8. A vehicle-like lane-changing control system that integrates predicted trajectories, characterized in that, include: An extraction module is constructed to build a set of expert trajectories for lane changing based on historical vehicle trajectory data. A set of candidate trajectories for lane changing is generated by perturbing the endpoint state of the expert trajectories and using polynomial interpolation. Extract trajectory features from the expert trajectory set and the candidate trajectory set for lane changing of vehicles, and construct trajectory feature vectors by combining environmental and future vehicle trajectory information; The learning module is used to learn a human-like lane-changing reward function by using the expert trajectory set as the real sample, the candidate trajectory set as the decision space, and the trajectory feature vector as the feature input, through the maximum entropy principle and feature matching mechanism. Based on the human-like lane-changing reward function, the candidate trajectory with the highest reward value is selected from the vehicle lane-changing candidate trajectory set as the human-like lane-changing trajectory. The human-like lane-changing reward function is used to reproduce expert behavior preferences. The optimization module is used to construct a multi-objective optimization function based on a human-like lane-changing trajectory. It optimizes the trajectory parameters of the multi-objective optimization function in the spatiotemporal dimension through piecewise nonlinear programming to generate the optimal human-like lane-changing trajectory. The optimization objectives of the multi-objective optimization function include human-likeness, comfort, lane-changing efficiency, and safety. The control module is used to output the optimal human-like lane-changing trajectory as the tracking trajectory to the vehicle's lateral and longitudinal controllers, so that the vehicle's lateral and longitudinal controllers can complete the human-like lane-changing control process based on the tracking trajectory.

9. A vehicle-like lane-changing control device that integrates predicted trajectories, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the steps of the vehicle-human lane-changing control method of any one of claims 1-7 when executing the computer program.

10. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is executed by the processor, it is used to implement the steps of the vehicle-human lane-changing control method for fusing predicted trajectories as described in any one of claims 1-7.

Images