Event-triggered MPC automatic driving vehicle collision avoidance method based on neural network

Abstract

The invention relates to an event triggering MPC automatic driving vehicle collision avoidance method based on a neural network, and belongs to the technical field of automatic driving vehicle collision avoidance. The method includes: acquiring whether an obstacle exists based on a sensing module; when the obstacle exists, judging a path planning trigger condition; when a path planning triggering condition is met, training a neural network based on the vehicle state data to obtain vehicle disturbance compensation, and adding the vehicle disturbance compensation to the vehicle dynamics model; and designing an MPC obstacle avoidance trajectory planner for the vehicle dynamics model based on a cost function and constraint conditions, and outputting a planning reference trajectory. According to the method, interference compensation is added in the vehicle nominal model, the modeling precision is improved, the triggering frequency is reduced, and the calculated amount is reduced. Compared with a traditional event triggering MPC method, the event triggering MPC based on the neural network can reduce the triggering frequency and reduce the tracking error.

Description

Technical Field

[0001] The present invention relates to the technical field of collision avoidance of autonomous driving vehicles, and in particular to an event-triggered MPC autonomous driving vehicle collision avoidance method based on a neural network. Background Art

[0002] In the field of autonomous driving, path planning and tracking have become a hot topic of research at home and abroad. Faced with complex traffic environments and roads, efficient collision avoidance strategies are particularly important for autonomous vehicles to complete specified tasks. Smart cars consider road conditions and vehicle dynamics to search for the optimal path, thereby adapting to different traffic scenarios and vehicle states. The motion planner can find a feasible collision-free path based on the vehicle's current position, target position, and surrounding environment information through algorithms and calculations to guide the object from the starting position to the destination. At the same time, the motion controller provides control force to enable it to travel along a predetermined trajectory to achieve collision avoidance driving tasks. According to different driving scenarios and needs, help the vehicle plan trajectories that adapt to different traffic environments in advance, solve vehicle obstacle avoidance problems to avoid potential collision risks, ensure the stability of collision avoidance, and meet the real-time requirements of autonomous driving vehicles.

[0003] The traditional MPC algorithm plans and tracks collision-free trajectories at a fixed frequency, which cannot adapt to the rapidly changing conditions in the actual driving environment. Especially in the case of rapid changes, there may be deviations between the planned path and the actual required path. Although the event-triggered MPC algorithm performs vehicle movement and planning at different frequencies, the generated control strategy cannot fully adapt to changes in the actual environment because the interference error is not considered in its nominal model. Therefore, it is necessary to combine real-time feedback to compensate for model uncertainty and dynamically adjust the planned path to make the generated control strategy more robust. Summary of the invention

[0004] The technical problems to be solved by the present invention are:

[0005] Although the traditional MPC algorithm can perform planning and tracking at a fixed frequency, it cannot adapt to changes in the environment in real time. The event-triggered MPC algorithm performs planning and control when the trigger conditions are met, which can respond to changes more flexibly, but does not consider interference errors. The generated control strategy has poor robustness and cannot effectively respond to the uncertainty of the actual environment. In order to avoid the shortcomings of the prior art, the present invention provides an event-triggered MPC autonomous driving vehicle collision avoidance method based on a neural network.

[0006] In order to solve the above technical problems, the technical solution adopted by the present invention is:

[0007] An event-triggered MPC autonomous driving vehicle collision avoidance method based on a neural network, characterized by comprising:

[0008] Determine whether there is an obstacle based on the sensor module;

[0009] When there are obstacles, the path planning triggering conditions are determined;

[0010] When the path planning triggering condition is met, a neural network is trained based on the vehicle state data to obtain vehicle disturbance compensation, and the vehicle disturbance compensation is added to the vehicle dynamics model;

[0011] An MPC obstacle avoidance trajectory planner is designed for the vehicle dynamics model based on the cost function and constraint conditions, and a planned reference trajectory is output.

[0012] A further technical solution of the present invention is as follows: the path planning triggering condition is determined as follows:

[0013] The vehicle tracking error is obtained, and the vehicle tracking error and the set threshold are judged. When the vehicle tracking error is greater than the set threshold, the path planning is triggered.

[0014] A further technical solution of the present invention is as follows: the vehicle tracking error is the error between the planned reference path of the vehicle and the actual path of the vehicle, specifically:

[0015]

[0016] Where Δ is the trigger level, O is the path tracking error weight matrix, which adjusts the contribution of errors in different directions to the trigger condition, and t k+1 ' is the trigger time, indicating that the trajectory tracking error reaches the trigger level, ρ is the upper bound of the disturbance, β is the trigger parameter, T is the length of the prediction time, I represents the neural network learning vector, is the largest eigenvalue of the matrix, is the state vector of the reference trajectory, A k is the system matrix, B k is the control matrix, D k represents the error matrix, is the control input vector, S, t k They are the end time and start time of the prediction respectively, and the start time is also the time when the event is triggered.

[0017] A further technical solution of the present invention: the neural network comprises an input layer, a hidden layer and an output layer;

[0018] The input vector of the input layer i(t) = [v x (t) v y (t) w(t) δ(t)F x (t)], where v x (t), v y (t), w(t), δ(t), F x(t) represent the vehicle longitudinal velocity, lateral velocity, angular velocity, front wheel steering angle and tire longitudinal force respectively;

[0019] The output data of the output layer is the difference between the updated predicted state and the actual measured state, which is used to train the weight parameters of the model at the next moment:

[0020]

[0021] in, They represent the longitudinal velocity, lateral velocity and angular velocity corresponding to the reference path respectively.

[0022] A further technical solution of the present invention is to select a Gaussian function as the activation function of the neural network for training, that is:

[0023]

[0024] Among them, c j is the center vector of the Gaussian basis function, indicating the center position of the function in the feature space; σ j is the width of the Gaussian basis function; the third output layer is the overall output of the neural network, and the predicted output of the network is calculated by the linear combination of the hidden layers:

[0025]

[0026] Among them, m is the number of hidden layer nodes, and the training error can be expressed as:

[0027]

[0028] During the training process, the total error is minimized by adjusting the weight parameters, and the updated weight matrix is ​​substituted into the disturbance compensation term of the vehicle motion model to improve the accuracy of trajectory prediction.

[0029] A further technical solution of the present invention is to add the vehicle disturbance compensation to the vehicle dynamics model to obtain a vehicle dynamics model with added vehicle disturbance compensation terms:

[0030]

[0031] where nn(·) is an additional neural network function used to estimate the nominal model modeling error, is the state vector derivative of the reference trajectory, is the state vector of the reference trajectory, is the control input vector.

[0032] A further technical solution of the present invention: The MPC obstacle avoidance trajectory planner is specifically:

[0033]

[0034]

[0035]

[0036]

[0037]

[0038] s∈[t k , t k +T]

[0039] in, is the control input sequence corresponding to the minimization of the cost function J, which drives the driving vehicle to track the expected path and avoid obstacles, and J is the total cost function.

[0040] A computer system, characterized in that it comprises: one or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the above-mentioned method.

[0041] A computer-readable storage medium, characterized in that it stores computer-executable instructions, which are used to implement the above method when executed.

[0042] A computer program product, characterized by comprising computer executable instructions, wherein the instructions are used to implement the above method when executed.

[0043] The beneficial effects of the present invention are:

[0044] The present invention provides an event-triggered MPC autonomous driving vehicle collision avoidance method based on a neural network. The event-triggered MPC method is adopted to ensure that the reference path is replanned only when the tracking error reaches the trigger level, so that the path planning and motion control are performed at different time intervals to improve the adaptability of trajectory tracking. The uncertainty of vehicle modeling is approximated by using a neural network. Interference compensation is added to the vehicle nominal model to improve the modeling accuracy, reduce the trigger frequency, and reduce the amount of calculation. Compared with the traditional event-triggered MPC method, the event-triggered MPC based on a neural network can reduce the number of triggers and reduce the tracking error.

[0045] 1. The present invention utilizes the learning ability of neural networks to compensate for modeling uncertainty, and introduces the training results of the neural network into the cost function calculation of the model predictive control algorithm, thereby dealing with unpredictable environmental interference and improving the accuracy of the model.

[0046] 2. The vehicle kinematics and dynamics models are introduced into the model predictive control algorithm, and the objective function of the collision avoidance strategy is designed. The vehicle state constraints and control input constraints considering actuator saturation under the collision avoidance strategy are formulated to ensure the feasibility of the planned trajectory.

[0047] 3. By formulating trigger rules, the reference path is replanned only at the event trigger time, and vehicle motion planning and control are executed at different frequencies, which improves the vehicle's adaptability to trajectory tracking errors and achieves precise control of the vehicle. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] The drawings are only for the purpose of illustrating particular embodiments and are not to be considered limiting of the present invention. Like reference symbols denote like components throughout the drawings.

[0049] Figure 1 Flow chart of the method of the present invention.

[0050] Figure 2 Vehicle plan motion diagram.

[0051] Figure 3 Diagram of the structure of a single hidden layer neural network.

[0052] Figure 4 Vehicle lateral collision avoidance constraints.

[0053] Figure 5 Neural network based event-triggered MPC.

[0054] Figure 6 Tracking error plot for event-triggered MPC based on a neural network.

[0055] Figure 7 Trajectory tracking based on traditional event-triggered MPC.

[0056] Figure 8 Tracking error plot for trajectory tracking based on traditional event-triggered MPC. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0058] The present invention provides an event-triggered MPC autonomous driving vehicle collision avoidance method based on a neural network, establishes a kinematic and dynamic model of the vehicle, considers driving safety and driving stability, formulates vehicle state constraints and control input constraints to ensure the feasibility of trajectory planning. A vehicle motion model is established based on a neural network to estimate the modeling error of the nominal model, handle unpredictable environmental interference, and improve the accuracy of the model. Trigger rules are formulated using path tracking errors, and the reference path is replanned only when the trigger condition is met. Vehicle motion and planning are performed at different frequencies, thereby improving the adaptability of trajectory tracking.

[0059] Includes the following:

[0060] 1. Vehicle kinematics model and dynamics model

[0061] First, the vehicle motion model is designed to solve the path planning and tracking problems, so that the designed motion controller can execute and track the reference collision-free trajectory. Figure 2 As shown, G represents the center of gravity of the vehicle, l f and l r represents the distance from the center of gravity of the vehicle to the front and rear wheels, δ is the front wheel steering angle, ψ is the vehicle yaw angle, and ψ r is the desired yaw angle. Then the vehicle kinematic equation can be expressed as:

[0062]

[0063] The vehicle dynamics equation is:

[0064]

[0065] Where m is the total mass of the vehicle, I z is the vertical moment of inertia of the vehicle. F xf (t) and F xr (t) are the longitudinal tire forces of the front and rear wheels, respectively, and F yf (t) and F yr (t) are the lateral tire forces of the front and rear wheels, respectively. This dynamic model can more accurately describe the basic characteristics of the vehicle when it is driving at high speed. Since vehicle motion control includes longitudinal adjustment and lateral tracking, in order to simplify the design of the controller, the vehicle front wheel steering angle and tire longitudinal force are used as control inputs, and the state space expression of the vehicle system is obtained:

[0066]

[0067] The above expression is a realistic model of vehicle motion, including indescribable external disturbances. k represents the error matrix. Under normal driving conditions, since the heading angle error is small, it is assumed that sinψ e ≈ψe , cosψ e ≈1, then the system matrix is ​​expressed as:

[0068]

[0069] The control matrix is ​​expressed as:

[0070]

[0071] Among them, C f and C r Represent the front and rear wheel cornering stiffness respectively.

[0072] 2. Interference error compensation based on neural network

[0073] Since the actual vehicle motion is disturbed by the uncertain environment, adding disturbance compensation to the vehicle dynamics model can reduce the prediction error and improve the model accuracy. For a nominal vehicle system without modeling error and external disturbance, the state space expression is:

[0074]

[0075] Design a vehicle motion model with disturbance compensation, and its state space expression is:

[0076]

[0077] Where nn(·) is an additional neural network function used to estimate the modeling error of the nominal model. Assuming that the vehicle model error is mainly caused by the dynamic part of the system equation, the input vector of the training neural network is extracted independently from the vehicle model, that is:

[0078] i(t)=[v x (t) v y (t) w(t) δ(t) F x (t)] (8)

[0079] Among them, v x (t), v y (t), w(t), δ(t), F x (t) represent the vehicle longitudinal velocity, lateral velocity, angular velocity, front wheel steering angle and tire longitudinal force respectively;

[0080] The difference between the updated predicted state and the actual measured state is used as the output layer data of the neural network online learning to train the weight parameters of the model at the next moment:

[0081]

[0082] In order to simplify the calculation, a single hidden layer neural network approximator is selected to deal with the uncertainty of structure and parameters. The network structure is as follows Figure 3 shown.

[0083] The present invention selects the Gaussian function as the activation function of the neural network, that is:

[0084]

[0085] Among them, c j is the center vector of the Gaussian basis function, indicating the center position of the function in the feature space. j is the width of the Gaussian basis function. The third output layer is the overall output of the neural network, and the predicted output of the network is calculated by the linear combination of the hidden layers:

[0086]

[0087] Among them, m is the number of hidden layer nodes, and the training error can be expressed as:

[0088]

[0089] During the training process, the total error is minimized by adjusting the weight parameters, and the updated weight matrix is ​​substituted into the disturbance compensation term of the vehicle motion model to improve the accuracy of trajectory prediction.

[0090] 3. Path planning trigger rule formulation

[0091] By designing trigger rules, it is possible to determine when to activate a new collision avoidance strategy, so that the frequency of path planning and control can be flexibly adjusted. The basic process is as follows:

[0092] 1. Define the triggering conditions for activating a new collision avoidance strategy based on the vehicle motion status information;

[0093] 2. When the trigger conditions are met, the path planning algorithm is started to find a safe and feasible path, avoid obstacles, and reach the target point within a given time range;

[0094] 3. Considering the dynamic constraints of the vehicle and environmental changes, the model predictive control algorithm is used to calculate the control force at each time step on the planned path and generate an adaptive control strategy;

[0095] 4. At each time step, the vehicle is controlled according to the calculated control force so that it moves along the planned trajectory;

[0096] 5. Set the next event trigger point. When the vehicle reaches this point, activate the new model predictive control algorithm for path planning and control, and repeat the above process until the vehicle reaches the destination.

[0097] Through the above steps, asynchronous execution of path planning is achieved. The present invention takes into account external interference and modeling errors, and uses path tracking errors to formulate trigger rules. The trigger time is defined as:

[0098]

[0099] Where Δ is the trigger level, and O is the path tracking error weight matrix, which adjusts the contribution of errors in different directions to the trigger condition. k+1 ' is the trigger time, indicating that the trajectory tracking error reaches the trigger level. The trigger level Δ can be expressed as:

[0100]

[0101] Where ρ is the upper bound of the disturbance and β is the trigger parameter. When the trigger condition is met within the prediction time range, the trigger time is t k+1 ', that is, the trajectory tracking error reaches the trigger level; in addition, it is necessary to generate a reference trajectory at the initial position and perform vehicle motion planning when the vehicle reaches the end of the planned trajectory. The length of the prediction time is defined as T. If the trigger rule is not satisfied within the prediction range, the trigger time is t k +T. The overall trigger condition expression is as follows:

[0102] t k+1 =min(t k+1 ',t k +T) (15)

[0103] 4. Collision avoidance controller design

[0104] In order to ensure driving safety, it is necessary to limit the vehicle's state to an allowable range so that the vehicle can change its path along the planned trajectory.

[0105] Vehicle restrictions are:

[0106]

[0107] The above constraints can be used to avoid possible collisions with obstacles. The above constraints can be organized as follows:

[0108]

[0109] The goal of the collision avoidance strategy is to find a collision-free path to the destination and calculate the control amount within the predicted horizon. The definition of the cost function needs to comprehensively consider the control cost and driving stability, and its expression is:

[0110]

[0111] Where R is the control force weight matrix; M is the control input change weight matrix; Q is the state deviation weight matrix; T is the prediction time domain length, and J is the total cost function.

[0112] Based on the vehicle motion model, combined with the cost function and constraints, the expression of the collision avoidance strategy is obtained as follows:

[0113]

[0114] By solving the above optimization problem, the control quantity of the planned path within the prediction range is obtained, so that the purpose of driving the vehicle to track the expected path and avoid obstacles is achieved.

[0115] According to the above theoretical description and formula derivation, a method for collision avoidance of an MPC autonomous driving vehicle triggered by an event based on a neural network is obtained according to an embodiment of the present invention, which specifically includes the following steps:

[0116] Step 1: Determine whether there is an obstacle based on the sensor module;

[0117] Step 2: When there is an obstacle, determine the path planning trigger condition;

[0118] Step 3: When the path planning trigger condition is met, a neural network is trained based on the vehicle state data to obtain vehicle disturbance compensation, and the vehicle disturbance compensation is added to the vehicle dynamics model;

[0119] Step 4: Design an MPC obstacle avoidance trajectory planner for the vehicle dynamics model based on the cost function and constraints, and output a planned reference trajectory.

[0120] 5. Simulation Experiment

[0121] In the simulation environment, considering the lane change driving scenario with road obstacles, the MPC algorithm based on neural network is used to generate the reference trajectory and the actual tracking path. When the tracking error reaches the threshold, the reference trajectory is generated. Or when the vehicle reaches the end of the planned trajectory, vehicle motion planning is also required. In the simulation, the prediction step size is set to 30 and the target lane centerline is set to y=4. The simulation results are shown in Figure 2. Figure 5 As shown:

[0122] As can be seen from the figure above, the algorithm plans a total of three reference paths. The first time, a reference trajectory is generated at the initial position, and the planned path can avoid obstacles; tracking errors will inevitably occur during the tracking process. Until the trigger condition is met, the motion planning is re-performed at the longitudinal position of 50m, and the corresponding control force is generated to drive the vehicle to track the new reference path with a smaller error. The third trigger point is activated at the longitudinal position of 108m, and the generated reference path guides the vehicle to drive smoothly on the center line of the target lane. The tracking error diagram is as follows:

[0123] exist Figure 6 In the figure, the path tracking error is kept within 0.15m, the lower limit of the tracking error is reached at the second trigger point (longitudinal position 50m), and the upper limit of the tracking error is reached at the third trigger point (longitudinal position 108m). Since the reference path is replanned after the trigger condition is met, the error is always kept within the allowable range. Compared with the MPC collision avoidance method proposed in the present invention, the traditional event-triggered MPC is used for planning and tracking in the same driving scenario. The tracking trajectory is shown in the figure below:

[0124] Depend on Figure 7 It can be seen that under event-triggered MPC, the planner triggers and generates a total of 5 reference paths, and the path after the vehicle stabilizes has a certain deviation from the center line of the target lane. This is because, without the neural network simulation uncertainty, the motion planner cannot make corresponding compensation adjustments based on the feedback of the actual tracking error.

[0125] Tracking error Figure 8 As shown, from Figure 8 It can be seen that this method has poor adaptability to tracking errors and large tracking errors. The event-triggered MPC obstacle avoidance strategy based on neural networks makes up for the shortcomings of this method.

[0126] The above description is only a specific implementation mode of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should be included in the protection scope of the present invention.

Claims

1. A neural network-based event-triggered MPC autonomous vehicle collision avoidance method, It is characterized in that include: Determine whether there is an obstacle based on the sensor module; When there are obstacles, the path planning triggering conditions are determined; When the path planning triggering condition is met, a neural network is trained based on the vehicle state data to obtain vehicle disturbance compensation, and the vehicle disturbance compensation is added to the vehicle dynamics model; An MPC obstacle avoidance trajectory planner is designed for the vehicle dynamics model based on the cost function and constraint conditions, and a planned reference trajectory is output.

2. According to the neural network-based event-triggered MPC autonomous driving vehicle collision avoidance method of claim 1, It is characterized in that The path planning triggering condition is determined as follows: The vehicle tracking error is obtained, and the vehicle tracking error and the set threshold are judged. When the vehicle tracking error is greater than the set threshold, the path planning is triggered.

3. According to the neural network-based event-triggered MPC autonomous driving vehicle collision avoidance method of claim 2, It is characterized in that The vehicle tracking error is the error between the planned reference path of the vehicle and the actual path of the vehicle, specifically: Where Δ is the trigger level, O is the path tracking error weight matrix, which adjusts the contribution of errors in different directions to the trigger condition, and t k+1 ' is the trigger time, indicating that the trajectory tracking error reaches the trigger level, ρ is the upper bound of the disturbance, β is the trigger parameter, T is the length of the prediction time, I represents the neural network learning vector, is the largest eigenvalue of the matrix, is the state vector of the reference trajectory, A k is the system matrix, B k is the control matrix, D k represents the error matrix, is the control input vector, S, t k They are the end time and start time of the prediction respectively, and the start time is also the time when the event is triggered.

4. According to the neural network-based event-triggered MPC autonomous driving vehicle collision avoidance method of claim 1, It is characterized in that The neural network includes an input layer, a hidden layer and an output layer; The input vector of the input layer i(t) = [v x (t)v y (t)w(t)δ(t)F x (t)], where v x (t), v y (t), w(t), δ(t), F x (t) represent the vehicle longitudinal velocity, lateral velocity, angular velocity, front wheel steering angle and tire longitudinal force respectively; The output data of the output layer is the difference between the updated predicted state and the actual measured state, which is used to train the weight parameters of the model at the next moment: in, They represent the longitudinal velocity, lateral velocity and angular velocity corresponding to the reference path respectively.

5. According to claim 4, an event-triggered MPC autonomous driving vehicle collision avoidance method based on a neural network, It is characterized in that The Gaussian function is selected as the activation function of the neural network for training, that is: Among them, c j is the center vector of the Gaussian basis function, indicating the center position of the function in the feature space; σ j is the width of the Gaussian basis function; the third output layer is the overall output of the neural network, and the predicted output of the network is calculated by the linear combination of the hidden layers: Among them, m is the number of hidden layer nodes, and the training error can be expressed as: During the training process, the total error is minimized by adjusting the weight parameters, and the updated weight matrix is ​​substituted into the disturbance compensation term of the vehicle motion model to improve the accuracy of trajectory prediction.

6. According to the neural network-based event-triggered MPC autonomous driving vehicle collision avoidance method of claim 1, It is characterized in that The vehicle disturbance compensation is added to the vehicle dynamics model to obtain the vehicle dynamics model with the vehicle disturbance compensation term added: where nn(·) is an additional neural network function used to estimate the nominal model modeling error, is the state vector derivative of the reference trajectory, is the state vector of the reference trajectory, is the control input vector.

7. According to the neural network-based event-triggered MPC autonomous driving vehicle collision avoidance method of claim 1, It is characterized in that The MPC obstacle avoidance trajectory planner is specifically: s∈[t k ,t k +T] in, is the control input sequence corresponding to the minimization of the cost function J, which drives the driving vehicle to track the expected path and avoid obstacles, and J is the total cost function.

8. A computer system, Features include: One or more processors, and a computer-readable storage medium for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors are enabled to implement the method of claim 1.

9. A computer-readable storage medium, Features Computer executable instructions are stored, and when the instructions are executed, they are used to implement the method of claim 1.

10. A computer program product, Features The invention comprises computer executable instructions, which are used to implement the method of claim 1 when being executed.

Images