Rear wheel active steering control method and system based on model prediction and reinforcement learning

Abstract

The invention provides a rear wheel active steering control method and system based on model prediction and reinforcement learning, and relates to the technical field of automobile control, and the method comprises the steps: inputting a front wheel steering angle, a vehicle speed, a yaw velocity and a side slip angle of a vehicle into a model prediction controller, and taking a given reference trajectory as a control target; the rolling optimization model predicts the output of the controller, and the optimal rear wheel steering angle control quantity is obtained through output; inputting vehicle state information and the optimal rear wheel steering angle control quantity into a deep reinforcement learning network, taking a multi-layer full-connection neural network as an action network of Actor in a deep deterministic strategy gradient reinforcement learning algorithm, giving an expected value, setting an observation state value and an award, and obtaining an optimal rear wheel steering angle control quantity; and a depth deterministic strategy gradient reinforcement learning algorithm is used for carrying out target tracking on the side slip angle and the yaw velocity, so that the yaw velocity and the side slip angle are close to expected values as much as possible, and optimal rear wheel steering control is realized.

Description

Technical Field

[0001] The present disclosure relates to the field of automobile control technology, and in particular to a rear-wheel active steering control method and system based on model prediction and reinforcement learning. Background Art

[0002] The statements in this section merely provide background information related to the present disclosure and do not necessarily constitute prior art.

[0003] Vehicle handling stability is increasingly valued as an important performance characteristic of modern vehicles. Vehicle active safety technology can assist the driver's control actions, minimizing or eliminating the probability of vehicle accidents. Therefore, effectively improving the vehicle's handling stability in dangerous situations has long been a key focus in vehicle research. Among these, the main technologies that affect vehicle handling stability are steering and stability control technologies, including front-wheel active steering, rear-wheel active steering, active four-wheel steering, servo steering, and vehicle stability control technologies. These technologies are primarily achieved through vehicle lateral force (steering control) and longitudinal force (driving and braking) control. Typical examples include vehicle stability control technology and rear-wheel active steering technology.

[0004] Rear-wheel active steering (RWS) is one method for achieving active vehicle safety. Its primary advantage lies in controlling the vehicle's slip angle, essentially maintaining it at zero during steering. This significantly improves the vehicle's dynamic response to steering wheel inputs and the transient response to yaw rate and lateral acceleration. Furthermore, for long-wheelbase vehicles, it significantly reduces the turning radius at low speeds, improving low-speed maneuverability. Existing RWS methods rely on simple front-to-rear wheel proportional control or feedback control based on the vehicle's yaw rate. These methods lack adaptive control capabilities based on the environment. Summary of the Invention

[0005] To solve the above problems, the present disclosure proposes a rear-wheel active steering control method and system based on model prediction and reinforcement learning. A simplified two-degree-of-freedom single-track model of the whole vehicle is used to design a model predictive controller. The dynamic state of the vehicle body and the corresponding output of the model predictive controller are obtained based on joint simulation. A multi-layer fully connected neural network is then used as the action network of the Actor in the reinforcement learning algorithm DDPG. The DDPG algorithm is used to perform trajectory tracking and heading angle tracking multi-condition learning in the CarSim platform. Finally, the DDPG intelligent agent algorithm trained based on multi-condition learning is used for multi-condition learning of a real vehicle, thereby achieving active rear-wheel steering control with optimal comprehensive performance.

[0006] According to some embodiments, the present disclosure adopts the following technical solutions:

[0007] The rear-wheel active steering control method based on model prediction and reinforcement learning includes:

[0008] Construct a two-degree-of-freedom vehicle monorail model and design a model predictive controller based on the two-degree-of-freedom vehicle monorail model;

[0009] The vehicle's front wheel steering angle, speed, yaw rate, and sideslip angle are input into the model predictive controller. The given reference trajectory is used as the control target, and the output of the model predictive controller is continuously optimized to obtain the optimal rear wheel steering angle control value.

[0010] The vehicle state information and the optimal rear wheel steering angle control amount are input into the deep reinforcement learning network. The multi-layer fully connected neural network is used as the action network of the Actor in the deep deterministic policy gradient reinforcement learning algorithm. Given the expected value, the observed state value and the reward are set. The deep deterministic policy gradient reinforcement learning algorithm is used to track the center of mass sideslip angle and yaw rate targets, so that the yaw rate and center of mass sideslip angle are as close to the expected value as possible to achieve optimal rear wheel steering control.

[0011] According to some embodiments, the present disclosure adopts the following technical solutions:

[0012] The rear-wheel active steering control system based on model prediction and reinforcement learning includes:

[0013] A model building module is used to build a two-degree-of-freedom vehicle monorail model and design a model predictive controller based on the two-degree-of-freedom vehicle monorail model;

[0014] The predictive control module is used to input the vehicle's front wheel angle, vehicle speed, yaw rate, and sideslip angle into the model predictive controller, use the given reference trajectory as the control target, and perform rolling optimization on the output of the model predictive controller to obtain the optimal rear wheel angle control value;

[0015] The active steering control module is used to input vehicle state information and the optimal rear wheel steering angle control amount into the deep reinforcement learning network. The multi-layer fully connected neural network is used as the action network of the Actor in the deep deterministic policy gradient reinforcement learning algorithm. Given the expected value, the observed state value and reward are set. The deep deterministic policy gradient reinforcement learning algorithm is used to track the center of mass sideslip angle and yaw rate targets, so that the yaw rate and center of mass sideslip angle are as close to the expected value as possible to achieve optimal rear wheel steering control.

[0016] According to some embodiments, the present disclosure adopts the following technical solutions:

[0017] A computer program product includes a computer program, which, when executed by a processor, implements the rear wheel active steering control method based on model prediction and reinforcement learning.

[0018] According to some embodiments, the present disclosure adopts the following technical solutions:

[0019] A non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by a processor, the rear wheel active steering control method based on model prediction and reinforcement learning is implemented.

[0020] According to some embodiments, the present disclosure adopts the following technical solutions:

[0021] An electronic device includes: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory to enable the electronic device to implement the rear wheel active steering control method based on model prediction and reinforcement learning.

[0022] Compared with the prior art, the present invention has the following beneficial effects:

[0023] The disclosed rear-wheel active steering control method based on model prediction and reinforcement learning takes the entire vehicle system as the research object. First, a simplified two-degree-of-freedom single-track model of the entire vehicle is used as the prediction model to design a model predictive controller. The dynamic state of the vehicle body and the corresponding output of the model predictive controller are obtained based on Simulink / CarSim joint simulation, and this is used as a data set to train a multi-layer fully connected neural network. The trained neural network is then used as the action network of the Actor in the reinforcement learning algorithm DDPG, and the DDPG algorithm is used to perform trajectory tracking, heading angle tracking and other task learning in the CarSim platform. Finally, the DDPG intelligent agent algorithm based on multi-working condition learning and training is subjected to real-vehicle multi-working condition learning and training, thereby achieving an active rear-wheel steering control system with optimal comprehensive performance. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings, which constitute a part of the present disclosure, are used to provide a further understanding of the present disclosure. The exemplary embodiments of the present disclosure and their descriptions are used to explain the present disclosure and do not constitute an improper limitation to the present disclosure.

[0025] Figure 1 Schematic diagram of the model predictive control structure of an embodiment of the present disclosure;

[0026] Figure 2 This is a structural diagram of the Actor Network of an embodiment of the present disclosure;

[0027] Figure 3 Critic Network structure diagram of the embodiment of the present disclosure

[0028] Figure 4This is a flow chart of the reinforcement learning algorithm of an embodiment of the present disclosure. DETAILED DESCRIPTION

[0029] The present disclosure will be further described below with reference to the accompanying drawings and embodiments.

[0030] It should be noted that the following detailed descriptions are illustrative and intended to provide further explanation of the present disclosure. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by those skilled in the art to which the present disclosure belongs.

[0031] It should be noted that the terms used herein are only for describing specific embodiments and are not intended to limit the exemplary embodiments according to the present disclosure. As used herein, unless the context clearly indicates otherwise, the singular form is intended to include the plural form. In addition, it should be understood that when the terms "comprise" and / or "include" are used in this specification, they indicate the presence of features, steps, operations, devices, components and / or combinations thereof.

[0032] Example 1

[0033] In one embodiment of the present disclosure, a rear wheel active steering control method based on model prediction and reinforcement learning is provided, comprising:

[0034] Step 1: Build a two-degree-of-freedom vehicle monorail model and design a model predictive controller based on the two-degree-of-freedom vehicle monorail model;

[0035] Step 2: Input the vehicle's front wheel angle, speed, yaw rate, and sideslip angle into the model predictive controller. Using the given reference trajectory as the control target, the model predictive controller's output is continuously optimized to obtain the optimal rear wheel angle control value.

[0036] Step 3: Input the vehicle state information and the optimal rear wheel steering angle control value into the deep reinforcement learning network, and use the multi-layer fully connected neural network as the action network of the Actor in the deep deterministic policy gradient reinforcement learning algorithm. Given the expected value, set the observed state value and reward, and use the deep deterministic policy gradient reinforcement learning algorithm to track the center of mass sideslip angle and yaw rate target, so that the yaw rate and center of mass sideslip angle are as close to the expected value as possible to achieve optimal rear wheel steering control.

[0037] As an embodiment, the specific implementation process of the rear wheel active steering control method based on model prediction and reinforcement learning disclosed in the present invention is as follows:

[0038] Step 1: Construct a two-degree-of-freedom vehicle monorail model and design a model predictive controller based on the two-degree-of-freedom vehicle monorail model;

[0039] Specifically, a two-degree-of-freedom vehicle monorail model is constructed, including:

[0040]

[0041] Let X(t) = [βγ] T ,u(t)=δ r ,ω(t)=δ f ,Y=γ

[0042] The dynamic model matrix can be obtained as follows:

[0043]

[0044] Where,

[0045]

[0046] Among them, k f 、k r are the front and rear wheel cornering stiffness respectively; L f , L r is the distance from the front axle / rear axle to the center of mass, m is the vehicle mass, v is the vehicle longitudinal speed, I z is the moment of inertia of the vehicle around the z-axis, β is the sideslip angle of the center of mass, gamma is the yaw angular velocity, δ f is the front wheel turning angle, δ r is the rear wheel turning angle.

[0047] Furthermore, a model predictive controller is designed based on a two-degree-of-freedom vehicle monorail model, including discretizing the continuity equation to obtain a new discrete state space model, which is:

[0048]

[0049] In the formula, vector x is the state vector, matrix A is the system matrix, matrix B is the control matrix, matrix F is the interference matrix, vector u is the control vector, vector ω is the interference vector, vector Y is the output vector, and matrix C is the output matrix.

[0050] A=(I+T s )A c ,B=T s B u ,F=T s B ω

[0051] Among them, T s is the sampling period of the system discretization, and I is the identity matrix.

[0052] make,

[0053]

[0054] A new discrete state space model can be obtained,

[0055]

[0056] in,

[0057]

[0058] Furthermore, in the predicted p-step time domain, the prediction equation is constructed as follows:

[0059] Y p (k+1|k)=S x Z(k)+S u ΔU(k)

[0060] in,

[0061]

[0062] Furthermore, given a reference quantity, the objective function is constructed, and the reference quantity is Taking the vehicle's front wheel angle, speed, yaw rate and sideslip angle as input, the given reference trajectory as the control target, rolling optimization and continuous correction of the output prediction value of the model predictive controller, the objective function is,

[0063] J=||Γ y (Y p (k+1|k)-R(k+1))|| 2

[0064] +‖Γ u ΔU(k)‖ 2

[0065] +ρε 2 +‖Γ u ΔU(k)‖ 2 +ρε 2 #(20)

[0066] Among them, Γ y , Γ u are the weighted matrices of reference and control variables, ρ is the weight coefficient, ε is the relaxation factor, and R(k+1) is the given control output reference sequence.

[0067]

[0068] The objective function and the corresponding control constraints are combined to form a quadratic programming problem, which is solved by the solver to obtain the optimal rear wheel steering angle control value δ r .

[0069] Step 2: Input the vehicle state information and the optimal rear wheel steering angle control value into the deep reinforcement learning network. Use the multi-layer fully connected neural network as the action network of the Actor in the deep deterministic policy gradient reinforcement learning algorithm. Given the expected value, set the observed state value and reward, and use the deep deterministic policy gradient reinforcement learning algorithm to track the center of mass sideslip angle and yaw rate target, so that the yaw rate and center of mass sideslip angle are as close to the expected value as possible to achieve optimal rear wheel steering control.

[0070] Specifically, step 1) collects vehicle status information based on Simulink / CarSim joint simulation data, and collects vehicle status information from the output of the MPC controller and the CarSim vehicle status information under scenarios of different vehicle speeds, road adhesion coefficients, and road profiles.

[0071] 2) A multi-layer, fully connected neural network was constructed in Matlab and pre-trained using the output of the MPC controller and the vehicle status information from CarSim. Pre-training process: Control sequences from the MPC controller were collected and used as a dataset to pre-train the Actor network used in deep reinforcement learning through deep learning (imitation learning).

[0072] Furthermore, we build a DDPG (Deep Deterministic Policy Gradient Reinforcement Learning) agent and learning environment, including:

[0073] The DDPG algorithm is a policy gradient-based deep reinforcement learning algorithm that can be used to solve problems in continuous action spaces. In continuous action spaces, agents can take actions of arbitrary values. Based on the Actor-Critic model, it uses a policy network μ(s; θ) to output decisions and a value network Q(s, a; ω) to score the decisions made by the policy network, thereby guiding its improvement. It also extends the DQN algorithm and uses experience replay and a target network to stabilize the training process.

[0074] In policy network training, DDPG uses deterministic policy gradient DPG to do Unbiased estimate of

[0075]

[0076] This gives us the algorithm for updating θ. Each time, a state is randomly extracted from the experience replay cache, denoted as s j , calculate a j =μ(s j ; θ), update θ using the gradient ascent formula:

[0077]

[0078] Here, β is the learning rate, which generally needs to be adjusted manually. Gradient ascent can increase the objective function J(θ), that is, let the value network give a higher score to the action taken by the decision network. In the training of the value network, similar to DQN, the TD algorithm is used to make the prediction of the value network closer to the TD target, and the prediction value of the target network closer to the true value function, that is,

[0079]

[0080] Its gradient is expressed as:

[0081]

[0082] The update formula of the value network is:

[0083]

[0084] In reinforcement learning, using a single "Q neural network" algorithm can lead to an unstable learning process. This is because the parameters of the value network are constantly modified during training and are also used in the calculation of the decision network. This mixed use will lead to large variance and instability during training, making it difficult for the algorithm to converge to the optimal solution. In order to provide stable estimation and improve training efficiency during training, the DDPG algorithm introduces two target networks, namely the target policy network μ(s; θ - ) and the target value network Q(s,a;ω - ), their update method adopts soft update, which is obtained by filtering the parameters of the main network. The detailed update formula is as follows.

[0085] θ - new ←τ·θ new +(1-τ)θ - now

[0086] ω - new ←τ·ω new +(1-τ)ω - now

[0087] Here, τ is the default parameter. Since the learning process becomes relatively slower and requires more time for training, more powerful hardware facilities are needed to meet the training requirements.

[0088] As an embodiment, the specific algorithm execution process is as follows:

[0089]

[0090] Among them, the observed states of the DDPG agent include: the difference between the center of mass sideslip angle and the expected value, the difference between the yaw angular velocity and the expected value, the center of mass sideslip angle error integral, the yaw angular velocity error integral, the center of mass sideslip angle measurement value, and the yaw angular velocity measurement value.

[0091] The reward design of the DDPG agent includes the following: the smaller the center of mass sideslip angle error and the smaller the yaw rate error, the greater the reward; a boundary is set, and penalties are imposed if the boundary is exceeded:

[0092]

[0093] Among them, the yaw rate error is: Center of mass sideslip angle error:

[0094] Where, K = m(l f / k r -l r / k f ) / (l f +l r ) 2 , K is the stability factor, an important parameter of the vehicle's steady-state response. L = l f +l r , μ is the road adhesion coefficient.

[0095] Perform training in the Simulink / CarSim environment until the desired effect is achieved, terminate the training, and save the model parameters.

[0096] Example 2

[0097] In one embodiment of the present disclosure, a rear wheel active steering control system based on model prediction and reinforcement learning is provided, comprising:

[0098] A model building module is used to build a two-degree-of-freedom vehicle monorail model and design a model predictive controller based on the two-degree-of-freedom vehicle monorail model;

[0099] The predictive control module is used to input the vehicle's front wheel angle, vehicle speed, yaw rate, and sideslip angle into the model predictive controller, use the given reference trajectory as the control target, and perform rolling optimization on the output of the model predictive controller to obtain the optimal rear wheel angle control value;

[0100] The active steering control module is used to input vehicle state information and the optimal rear wheel steering angle control amount into the deep reinforcement learning network. The multi-layer fully connected neural network is used as the action network of the Actor in the deep deterministic policy gradient reinforcement learning algorithm. Given the expected value, the observed state value and reward are set. The deep deterministic policy gradient reinforcement learning algorithm is used to track the center of mass sideslip angle and yaw rate targets, so that the yaw rate and center of mass sideslip angle are as close to the expected value as possible to achieve optimal rear wheel steering control.

[0101] Example 3

[0102] In one embodiment of the present disclosure, a computer program product is provided, including a computer program. When the computer program is executed by a processor, the computer program implements the rear wheel active steering control method based on model prediction and reinforcement learning.

[0103] Example 4

[0104] In one embodiment of the present disclosure, a non-transitory computer-readable storage medium is provided, which is used to store computer instructions. When the computer instructions are executed by a processor, the rear-wheel active steering control method based on model prediction and reinforcement learning is implemented.

[0105] Example 5

[0106] In one embodiment of the present disclosure, an electronic device is provided, comprising: a processor, a memory, and a computer program; wherein the processor is connected to the memory, and the computer program is stored in the memory. When the electronic device is running, the processor executes the computer program stored in the memory, so that the electronic device executes the rear-wheel active steering control method based on model prediction and reinforcement learning.

[0107] The present disclosure is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present disclosure. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0108] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0109] Although the above describes the specific implementation methods of the present disclosure in conjunction with the accompanying drawings, it is not intended to limit the scope of protection of the present disclosure. Those skilled in the art should understand that on the basis of the technical solution of the present disclosure, various modifications or variations that can be made by those skilled in the art without creative work are still within the scope of protection of the present disclosure.

Claims

1. A rear wheel active steering control method based on model prediction and reinforcement learning, characterized in that: include: Construct a two-degree-of-freedom vehicle monorail model and design a model predictive controller based on the two-degree-of-freedom vehicle monorail model; The front wheel steering angle, vehicle speed, yaw rate and sideslip angle of the vehicle are input into the model predictive controller, and the given reference trajectory is used as the control target. The output of the model predictive controller is optimized in a rolling manner to obtain the optimal rear wheel steering angle control value. The vehicle state information and the optimal rear wheel steering angle control amount are input into the deep reinforcement learning network, and the multi-layer fully connected neural network is used as the action network of the Actor in the deep deterministic policy gradient reinforcement learning algorithm. Given the expected value, the observed state value and the reward are set, and the deep deterministic policy gradient reinforcement learning algorithm is used to track the center of mass sideslip angle and yaw rate target, so that the yaw rate and center of mass sideslip angle are as close to the expected value as possible to achieve the optimal rear wheel steering control.

2. The rear wheel active steering control method based on model prediction and reinforcement learning as claimed in claim 1, characterized in that: Construct a two-degree-of-freedom vehicle monorail model, including: Take X(t)=[βγ] T ,u(t)=δ r ,δ(t)=δ f ,Y=γ; The kinetic model matrix is ​​as follows: In the formula, Among them, k f , k r are the front and rear wheel cornering stiffness respectively; L f , L r is the distance from the front axle / rear axle to the center of mass, m is the vehicle mass, v is the vehicle longitudinal velocity, I z is the moment of inertia of the vehicle around the z-axis, β is the sideslip angle of the center of mass, gamma is the yaw rate, δ f is the front wheel turning angle, δ r is the rear wheel turning angle.

3. The rear wheel active steering control method based on model prediction and reinforcement learning as claimed in claim 1, characterized in that: The model predictive controller is designed based on a two-degree-of-freedom vehicle single track model, including: discretizing the continuous equation to obtain a new discrete state space model. In the predicted p-step time domain, the prediction equation is: Y p (k+1|k)=S x Z(k)+S u ΔU(k)。 4. The rear wheel active steering control method based on model prediction and reinforcement learning as claimed in claim 1, characterized in that: The front wheel steering angle, vehicle speed, yaw rate and sideslip angle of the vehicle are input into the model predictive controller. The given reference trajectory is used as the control target. The output of the model predictive controller is optimized in a rolling manner to obtain the optimal rear wheel steering angle control value, including: Set the reference value to β d =0, The objective function and the corresponding control constraints are combined to form a quadratic programming problem, which is solved by the solver to obtain the optimal rear wheel steering increment δ r .

5. The rear wheel active steering control method based on model prediction and reinforcement learning as claimed in claim 1, characterized in that: The observed states of the action network agent of the Actor in the deep deterministic policy gradient reinforcement learning algorithm include: the difference between the center of mass sideslip angle and the expected value, the difference between the yaw angular velocity and the expected value, the center of mass sideslip angle error integral, the yaw angular velocity error integral, the center of mass sideslip angle measurement value and the yaw angular velocity measurement value; the reward is set so that the smaller the center of mass sideslip angle error and the smaller the yaw angular velocity error, the greater the reward; set boundaries and give penalties if the boundaries are exceeded.

6. The rear wheel active steering control method based on model prediction and reinforcement learning as claimed in claim 5, characterized in that: The deep deterministic policy gradient reinforcement learning algorithm is based on Actor-Critic, uses a policy network for decision output, uses a value network to score the decisions made by the policy network, uses experience replay and a target network, and uses the TD algorithm to make the prediction of the value network closer to the target and the prediction of the target network closer to the true value function.

7. Rear wheel active steering control system based on model prediction and reinforcement learning, characterized in that: include: A model building module, used to build a two-degree-of-freedom vehicle monorail model and design a model predictive controller based on the two-degree-of-freedom vehicle monorail model; The predictive control module is used to input the front wheel steering angle, vehicle speed, yaw rate and sideslip angle of the vehicle into the model predictive controller, take the given reference trajectory as the control target, and perform rolling optimization on the output of the model predictive controller to obtain the optimal rear wheel steering angle control value; The active steering control module is used to input vehicle state information and the optimal rear wheel steering angle control amount into the deep reinforcement learning network, use the multi-layer fully connected neural network as the action network of the Actor in the deep deterministic policy gradient reinforcement learning algorithm, give the expected value, set the observed state value and reward, and use the deep deterministic policy gradient reinforcement learning algorithm to track the center of mass sideslip angle and yaw rate target, so that the yaw rate and center of mass sideslip angle are as close to the expected value as possible to achieve the optimal rear wheel steering control.

8. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the rear wheel active steering control method based on model prediction and reinforcement learning as described in any one of claims 1 to 6 is implemented.

9. A non-transitory computer-readable storage medium, characterized in that: The non-transitory computer-readable storage medium is used to store computer instructions. When the computer instructions are executed by the processor, the rear-wheel active steering control method based on model prediction and reinforcement learning as described in any one of claims 1 to 6 is implemented.

10. An electronic device, characterized in that: include: A processor, a memory and a computer program; wherein the processor is connected to the memory, the computer program is stored in the memory, and when the electronic device is running, the processor executes the computer program stored in the memory so that the electronic device executes the rear-wheel active steering control method based on model prediction and reinforcement learning as described in any one of claims 1-6.

Images