Co-DMPC-based cooperative control method for autonomous vehicle chassis multi-agent system, controller, and storage medium

Abstract

Disclosed in the present invention are a Co-DMPC-based cooperative control method for an autonomous vehicle chassis multi-agent system, a controller, and a storage medium. In the present invention, a distributed state space equation having state coupling and control input coupling features is established, reflecting the mutual coupling between multiple agents, and taking into comprehensive account the influence of states and control inputs of neighbor agents, thus improving the accuracy of prediction of a future state sequence by a local agent; and the meanings of predicted trajectories, assumed trajectories and optimal trajectories of a state and a control input and conversion modes thereof are designed, thus providing a communication basis for information interaction between the agents. In order to coordinate global performance indices of a vehicle, a local agent optimization problem that accounts for cost coupling is established, and the influence of a cooperative relationship on the control effect is quantitatively analyzed by means of adaptive weight coefficients. Moreover, in the present invention, a method is used that involves performing multiple instances of iterative solving per unit sampling time, and an iterative error is used to balance the solution accuracy and solution efficiency of a controller.

Description

A Co-DMPC-based automatic driving vehicle chassis multi-agent system cooperative control method, controller and storage medium TECHNICAL FIELD

[0001] The present application relates to the field of automatic driving vehicle chassis distributed control, in particular to a chassis distributed control architecture design and control algorithm research based on multi-agent theory. BACKGROUND

[0002] Under the development trend of vehicle intelligence, electrification and drive-by-wire, higher requirements are put forward for the comfort, safety and functionality of the chassis system. More and more active / passive safety technologies have been applied, such as active front wheel steering (AFS), active rear wheel steering (ARS), direct yaw moment control (DYC) and other technologies. The development of more safety functions will inevitably lead to conflicts between different control objectives, such as the influence of DYC on longitudinal speed, the cooperation of ARS and front wheel steering direction, etc. Therefore, current researchers mainly use centralized control methods and decentralized control methods to coordinate the performance objectives between different systems.

[0003] For the centralized control method, the output of the whole system is finally fed back to the central controller, so the optimal control input of the whole system can be obtained. This method is relatively mature, and it can coordinate the complex coupling problems between multiple subsystems, but with the complication of chassis functions, the centralized control method will inevitably lead to an increase in solving difficulty, computational burden and high-performance controller cost. And the scalability of the centralized method is poor, when the controller fails, the whole system will not work normally. Unlike the centralized control method, the decentralized control method divides the whole system into several subsystems, each subsystem has a separate controller. Although there is a coupling relationship between the subsystems, the controllers are independent of each other and there is no information exchange between them, by ignoring the interaction between the controllers, the complexity and computational burden of the system are reduced. In addition, the decentralized method does not cause the entire system to be paralyzed when a single controller fails, and still retains a certain controllable ability, so it has good scalability and strong risk resistance, but in the strong coupling condition, it lacks means to handle the coupling state information.

[0004] And the distributed control method is a powerful tool for dealing with complex large systems, which has developed rapidly in the field of power grids, unmanned aerial vehicle formation, etc. in recent years. Compared with centralized and decentralized methods, it allows communication and information sharing between controllers, so that the coupling and coordination between subsystems can be fully considered, and the global optimal goal can be achieved. In addition, its strong scalability and high reliability meet the control requirements of multi-agent, so it is a powerful means to solve the coupling cooperative control.

[0005] Therefore, aiming at the contradictions and deficiencies existing above, the application proposes a collaborative distributed model predictive control (Co-DMPC) architecture based on a multi-agent system (MAS), iteratively solves the predicted state trajectory and the predicted control trajectory of each agent, and exchanges the prediction information with each other, thereby improving the control accuracy of the controller and reducing the operation time of the controller.

[0006] SUMMARY

[0007] To solve the above technical problems, the application provides a vehicle chassis collaborative control method, controller and storage device based on MAS for an automatic driving car with four-wheel drive and four-wheel steering function, realizes vehicle handling stability under complex conditions, and solves the deficiencies of the prior art.

[0008] S1: Establish a two-degree-of-freedom vehicle model of four-wheel steering + DYC. In the modeling process, the effect of the suspension is ignored. The model considers that the vehicle body only moves in a plane parallel to the ground, assumes that the turning angles of the left and right wheels on the same axle are the same, and regards the longitudinal velocity of the vehicle as constant. The model is shown in Figure 1. The simplified model is as follows:

[0009] In the formula, v x is the longitudinal velocity, v y is the lateral velocity, γ is the yaw rate, m s is the vehicle mass, C αf and C αr are the front and rear wheel cornering stiffness, l f and l r are the distances from the front and rear wheel centers to the center of mass, δ f and δ r are the front and rear wheel turning angles, I z is the rotational inertia of the vehicle rotating around the z-axis, and M z is the direct yaw moment.

[0010] Combined with the path tracking model, the following dynamic equation under the centralized control strategy can be obtained:

[0011] In the formula, Y is the lateral displacement of the vehicle, is the yaw angle.

[0012] S2: Establish a distributed state space equation. According to the different control variables, the centralized state space equation of the vehicle is divided into two agents, namely the steering agent and the DYC agent. For the steering agent, the control variables are δ f and δ r , and the state variables are v y and Y. The state space equation is:

[0013] where,

[0014] For DYC agents, the control variable is M z , the state variable it tracks is γ and The state space equation is:

[0015] where,

[0016] S3: Construct the collaborative control framework, as shown in Fig. 2. In the input layer, the vehicle model calculates the reference yaw rate γ ref and the reference lateral velocity v yref according to the deviation of the current state of the vehicle and the target path. The collaborative control layer decomposes the traditional chassis centralized control method into a distributed control system composed of multiple controllers, which respectively realize the tracking of the respective desired target. Through the state coupling space equation established in S2, the Co-DMPC algorithm is used to select the cost function with global performance target. In the unit sampling time, the iterative update method is used to enable each agent to independently calculate its own local optimization problem and exchange the calculation results with each other multiple times. By reasonably setting the iteration termination condition, the whole system can finally obtain the global optimal solution. The torque distribution layer optimizes the distribution of the driving / braking torque of the four wheels according to the total driving force calculated by the PID speed controller and the direct yaw moment calculated by the collaborative control layer. Finally, the control variables are applied to the four wheels through the steering actuator and the hub motor to realize accurate control of the vehicle.

[0017] S4: Establish an information transmission protocol. In all agents, the same prediction time domain N p and control time domain N c are used. For the i-th agent, in the control time domain [t, t+N c ], the control variable at time t is predicted at time t+k, denoted as u i (k|t). At the same time, the following three groups of control trajectories are defined:

[0018] (1) Predicted control trajectory;

[0019] (2) Optimal control trajectory;

[0020] (3) Assumed control trajectory.

[0021] where, represents the control variable predicted by the controller according to the state at the current time and the control variable at the previous time. The optimal solution obtained by the representative controller after solving the local optimization problem. The assumed control delivered by the representative agent i to its neighbors, which is obtained by shifting the predicted control by one sample period.

[0022] Within the prediction horizon [t, t+N p ], the state at time t+k is predicted at time t and denoted as x i (k|t). Three groups of state trajectories are defined as follows, which have the same meaning as the control trajectories:

[0023] (1) Predicted state trajectory;

[0024] (2) Optimal state trajectory;

[0025] (3) Assumed state trajectory.

[0026] The relationship among the predicted trajectory, the optimal trajectory and the assumed trajectory is shown in Fig. 3.

[0027] At time t, agent i utilizes the predicted state at future time k and the assumed state from its neighbors and the assumed control to solve the distributed optimization problem. When k = 0, where x i (t) and x j (t) are the states that can be directly measured by the vehicle at the current time. After obtaining the optimal solution state and the optimal solution control , the first of the control trajectory is applied to the actual control. The optimal state trajectory and the optimal control trajectory are recursively propagated for one step to obtain the assumed trajectory. That is, let:

[0028] At this time, it should be noted that the length of the control trajectory is N c -1, and the last item is padded, i.e. The state of the vehicle at the next time without control is

[0029] S5: Derivation of the local agent state equation. According to the characteristics of the state coupling of the multi-agent system, the local continuous linear state equation can be written as:

[0030] where m is the number of agents. Discretize the above equation as:

[0031] where:

[0032] By recursion in time domain The expression is:

[0033] The above recursive process can be written in the form of the following augmented matrix:

[0034] In the formula:

[0035] S6: Solving local optimization problems. The Co-DMPC algorithm is proposed to solve the parallel computing problem between subsystems with strong coupling characteristics, and is also applicable to the coordination and cooperation between sub-agents in multi-agent systems. By exchanging state information and control sequences, the global optimization problem can be solved. The system has good scalability, which facilitates the "plug and play" function of the vehicle chassis system, and reduces the software and hardware matching cost between different functional vehicle models.

[0036] For agent i, without considering the global performance index, i.e. non-collaborative distributed, the local optimization problem can be designed as follows:

[0037] The equation is constrained as follows:

[0038] In the formula, w i (t+k) is the state reference value of agent i, W i is the reference state sequence composed of w i (t+k). Q i and R i are state weight coefficients and control weight coefficients, respectively, and Specifically: Q i is the penalty for the deviation of the actual trajectory from the expected trajectory, reflecting the vehicle's expectation to follow its expected trajectory; R i is the penalty for the control input, reflecting the vehicle's expectation to travel with lower control energy. Constraint (17) reflects the vehicle's dynamics constraints. Constraints (18)-(19) require the initial state of the vehicle's prediction horizon to be equal to the vehicle's actual state. Constraint (20) is the input constraint of the agent, which ensures that the optimization result does not exceed the action range of the actuator.

[0039] For Co-DMPC algorithm, the difference between it and non-cooperative distributed model predictive control algorithm mainly lies in whether there is a global performance index. Each agent has a strong dependence on the state or control input of neighbor agent, so Co-DMPC needs strong communication capability. In addition, in the description of the optimization problem, there is cost coupling between the agent using Co-DMPC algorithm and the neighbor agent. Then the optimization problem considering the global performance index can be expressed as:

[0040] In the formula, λ i is the coupling coefficient of the cost function between different agents, and The constraint conditions are the same as formulas (17)-(20). In addition, in order to improve the operation efficiency of the controller, the optimization problem is converted into the form of quadratic programming:

[0041] In the formula, const is a constant term that does not affect the optimization result, and

[0042] After completing a single optimization calculation, the local agent only gets a local optimal solution and cannot achieve precise control of the vehicle. In order to reduce the error of optimization, the application adopts a cyclic iterative updating method, and in a single control period, the optimal control sequence and the optimal state sequence after optimization are exchanged multiple times, so as to realize the solution of the global optimum. At the same time, it should be noted that the use of iterative updating method actually increases the calculation burden of the controller and the occupancy rate of the communication network. Therefore, reasonable setting of the iteration termination condition is an important factor to ensure the real-time performance of the controller.

[0043] The application designs two termination conditions for the above requirements. First, let the error of the optimization results of adjacent two iterations be , which represents the optimization result of agent i in the nth iteration. When the iteration error Err i is less than the threshold value ε i , it is considered that the optimization result of the controller has converged to the global optimal solution, at which time the iteration is ended and the next step is performed. If the iteration error of the controller cannot always converge to ε i , a maximum iteration number p max needs to be set to prevent the controller from performing invalid calculation.

[0044] S7: Direct yaw moment distribution. After the DYC agent calculates the direct yaw moment of the vehicle, it needs to be reasonably distributed to the four wheels according to the state of the vehicle to avoid the phenomenon of tire force saturation. At the same time, the distributed torque also needs to meet the requirements of driving the vehicle forward and tracking the target speed. First, according to the PID algorithm, the total driving force equation of the vehicle is constructed:

[0045] where e(t) = v x -v xref Secondly, consider the following optimization problem

[0046] Since the steering agent has calculated the four-wheel steering angles at this moment, i.e., the tire lateral forces F yij , i = fl, fr, rl, rr are constants, which have no effect on the solution of the above optimization problem. Therefore, equation (25) can be written as

[0047] The above equation is subject to the following constraints:

[0048] where constraint (27) represents that the total driving force allocated to the four wheels should meet the requirements of the speed controller. Constraint (28) represents that the yaw moment formed by the longitudinal tire forces should meet the calculation results of the DYC agent. Constraint (29) represents that the size of the driving force should be within the range of the actuator. Where T min is the minimum driving torque (i.e., braking torque), T max is the maximum driving torque, and T max = -T min . The solution of equation (26) in MATLAB using the fmincon command can obtain the four-wheel driving / braking torque.

[0049] S8: Send the calculation results of the steering agent and the DYC agent to the actuator.

[0050] Advantages of the present application:

[0051] 1. The present application establishes a distributed state space equation with state coupling and control input coupling characteristics, and clarifies the mutual coupling mechanism between multiple agents. The influence of the state and control input of the neighbor agent is comprehensively considered, which improves the accuracy of the local agent in predicting the future state sequence;

[0052] 2. The present application designs the meanings and conversion methods of the predicted trajectory, the assumed trajectory and the optimal trajectory of the state and the control input, which provides a communication basis for the information interaction between agents. In order to coordinate the global performance index of the vehicle, a local agent optimization problem considering cost coupling is established, and the influence of the cooperative relationship on the control effect is quantitatively analyzed through an adaptive weight coefficient

[0053] 3. The present application uses a multiple iteration solution method within a unit sampling time, and uses the iteration error to balance the solution accuracy and solution efficiency of the controller. BRIEF DESCRIPTION OF DRAWINGS

[0054] Figure 1 is a diagram of a two-degree-of-freedom model of a vehicle considering rear wheel steering and direct yaw moment.

[0055] Figure 2 is a diagram of a multi-agent system control architecture.

[0056] Figure 3 is a diagram of an interaction information transformation architecture. DETAILED DESCRIPTION

[0057] For the convenience of those skilled in the art, the present application is further described below in conjunction with embodiments and drawings, and the content mentioned in the embodiments is not a limitation of the present application.

[0058] Step one: Establish a two-degree-of-freedom vehicle model of four-wheel steering + DYC. In the modeling process, the effect of the suspension is ignored. The model considers that the vehicle body only makes planar motion parallel to the ground, assumes that the turning angles of the left and right wheels on the same axle are the same, and considers the longitudinal velocity of the vehicle as constant. The model is shown in Figure 1. The simplified model is shown in the following formula:

[0059] In the formula, v x is the longitudinal velocity, v y is the lateral velocity, γ is the yaw angular velocity, m s is the vehicle mass, C αf and C αr are the front and rear wheel cornering stiffness, l f and l r are the distances from the front and rear wheel centers to the center of mass, δ f and δ r are the front and rear wheel turning angles, I z is the rotational inertia of the vehicle rotating around the z-axis, and M z is the direct yaw moment.

[0060] The following dynamics equation under the centralized control strategy can be obtained in combination with the path tracking model:

[0061] In the formula, Y is the lateral displacement of the vehicle, is the yaw angle.

[0062] Step two: Establish a distributed state space equation. According to the different control quantities, the centralized state space equation of the vehicle is divided into two agents, namely a steering agent and a DYC agent. For the steering agent, the control quantities are δ f and δ r , and the state quantities are v y and Y. The state space equation is:

[0063] In the formula

[0064] For DYC agents, the control variable is M z , and the state variable tracked is γ , and the state space equation is:

[0065] where

[0066] Step three: build a collaborative control framework, as shown in Figure 2, including four layers: (1) in the input layer, the vehicle model calculates the reference yaw rate γ ref and the reference lateral velocity v yref according to the deviation of the current state of the vehicle and the target path. (2) The collaborative control layer decomposes the traditional chassis centralized control method into a distributed control system composed of multiple controllers, which respectively realize the tracking of the respective desired target. Through the state coupling space equation established by S2, the Co-DMPC algorithm is adopted, and the cost function with global performance target is selected. In the unit sampling time, the iterative update method is adopted, so that each agent can independently calculate its own optimization problem, and the calculation results of each other are exchanged multiple times. By reasonably setting the iteration termination condition, the whole system can finally get the global optimal solution.(3) The torque allocation layer optimizes the allocation of four-wheel drive / brake torque according to the total driving force calculated by the PID speed controller and the direct yaw moment calculated by the upper layer.(4) The execution layer applies the control variable to the four wheels through the steering actuator and the hub motor to realize accurate control of the vehicle.

[0067] Step four: establish an information transmission protocol. In all agents, the same prediction horizon N p and control horizon N c are used. For the i-th agent, in the control horizon [t, t+N c ], the control variable at time t predicted at time t+k is denoted as u i (k|t). At the same time, the following three groups of control trajectories are defined:

[0068] (1) Predicted control trajectory;

[0069] (2) Optimal control trajectory;

[0070] (3) Assumed control trajectory.

[0071] where represents the controller predicting the control variable at future time according to the current state and the control variable at the previous time. represents the optimal solution obtained by the controller after solving the local optimal problem. The hypothetical control quantity passed by agent i to neighboring agents is composed of the predicted control quantity. This is obtained by shifting the entire trajectory by one sampling period T, which means moving the entire trajectory towards a future time, so that the element that was originally at time t=0 becomes the element at time t=1.

[0072] In the prediction time domain [t, t+N p Within [the context], the state variable predicted at time t+k is denoted as x. i (k|t). Simultaneously, the following three sets of state trajectories are defined, their meanings being similar to those of the control trajectories:

[0073] (1) Predict the state trajectory;

[0074] (2) Optimal state trajectory;

[0075] (3) Assume a state trajectory.

[0076] The relationship between the predicted trajectory, the optimal trajectory, and the hypothetical trajectory is shown in Figure 3.

[0077] Agent i uses the predicted state of future times at time t. Hypothetical state from neighboring agent j With assumption control Solve the distributed optimization problem. When k=0, Where, x i (t) and x j (t) represents the currently measurable state of the vehicle. The optimal solution state is obtained. With optimal solution control Then, control the first trajectory This is applied to actual control. The optimal state trajectory and optimal control trajectory are recursively compared step-by-step to obtain the hypothetical trajectory. That is, let:

[0078] At this point, it should be noted that the length of the control trajectory is N. c -1, to complete the last item, let The vehicle's state at the next moment is as follows: A ii It is the state matrix of agent i.

[0079] Step 5: Derivation of the Local Agent State Equation. Based on the characteristics of state coupling in multi-agent systems, the local continuous linear state equation can be written as:

[0080] In the formula, m is the number of agents. Discretizing the above formula, we get:

[0081] where

[0082] By recursion, we have The expression of in time domain is

[0083] The above recursive process can be written in the form of the following augmented matrix:

[0084] where

[0085] Step 6: Solve the local optimization problem. For agent i, without considering the global performance index, i.e. non-collaborative distributed, its local optimization problem can be designed as follows:

[0086] denotes the cost function considering only the local control performance.

[0087] The equation is subject to the following constraints:

[0088] where, U ii is the set of constraint conditions of control input, w i (t+k) is the state reference value of agent i, W i is the reference state sequence composed of w i (t+k). Q i and R i are state weight coefficient and control weight coefficient respectively, and Specifically: Q i is the penalty for the difference between the actual trajectory and the expected trajectory, reflecting the vehicle's expectation to follow its own expected trajectory; R i is the penalty for the control input, reflecting the vehicle's expectation to travel with lower control energy. Constraint (46) reflects the vehicle's dynamics constraints. Constraints (47)-(48) require the initial state of the vehicle's prediction horizon to be equal to the vehicle's actual state. Constraint (49) is the input constraint of the agent, which ensures that the optimization result does not exceed the action range of the actuator.

[0089] For Co-DMPC algorithm, the difference between it and non-cooperative distributed model predictive control algorithm mainly lies in whether there is a global performance index. Each agent has a strong dependence on the state or control input of neighbor agent, so Co-DMPC needs strong communication ability. In addition, when describing the optimization problem, there is cost coupling between the agent using Co-DMPC algorithm and the neighbor agent. Then the optimization problem considering the global performance index can be expressed as:

[0090] In the formula, represents the cost function considering the global performance, λ i , λ j is the coupling coefficient of the cost function between different agents, and The constraint condition is the same as formula (46)-(49). In addition, in order to improve the operation efficiency of the controller, the optimization problem is converted into the form of quadratic programming:

[0091] In the formula, const is a constant term that does not affect the optimization result, and

[0092] After completing a single optimization calculation, the local agent only gets a local optimal solution, which cannot realize accurate control of the vehicle. In order to reduce the error of optimization, the application adopts a cyclic iterative updating method, and in a single control period, the optimal control sequence and the optimal state sequence after optimization are exchanged multiple times, so as to realize the solution of the global optimal. At the same time, it should be noted that the use of iterative updating method actually increases the calculation burden of the controller and the occupancy rate of the communication network. Therefore, reasonable setting of the iteration termination condition is an important factor to ensure the real-time performance of the controller.

[0093] The application designs two termination conditions for the above requirements. First, the error of the optimization results of adjacent two iterations is denoted as , which represents the optimization result of agent i in the nth iteration. When the iteration error Err i is less than the threshold value ε i (i=1 represents the iteration error of the steering agent, and i=2 represents the iteration error of the DYC agent), it is considered that the optimization result of the controller has converged to the global optimal solution, and the iteration is ended at this time. If the iteration error of the controller cannot converge to ε i , a maximum iteration number p max needs to be set to prevent the controller from performing invalid calculation.

[0094] Step seven: Direct yaw moment distribution. After the DYC agent calculates the direct yaw moment, it needs to distribute it to the four wheels according to the state of the vehicle to avoid the phenomenon of tire force saturation. At the same time, the distribution torque must meet the requirements of vehicle driving forward and tracking the target speed.

[0095] First, according to the PID algorithm, the total driving force equation of the vehicle is constructed:

[0096] In the formula, K p represents the proportional coefficient, K i represents the integral coefficient, and K d represents the differential coefficient, e(t) = v x -v xref .

[0097] Second, consider the following optimization problem:

[0098] In the formula, J represents the abbreviation of the tire force distribution optimization problem, F xij represents the longitudinal force of the left front wheel, right front wheel, left rear wheel and right rear wheel, F zij represents the vertical force of the left front wheel, right front wheel, left rear wheel and right rear wheel, F yij represents the four tire lateral forces, and ij = fl, fr, rl, rr respectively represent the left front wheel, right front wheel, left rear wheel and right rear wheel. μ is the adhesion coefficient. Since the steering agent has calculated the steering angle of the four wheels at this time, the tire lateral force F yij is a constant, which does not affect the solution of the above optimization problem. Therefore, equation (54) can be written as:

[0099] The above formula is subject to the following constraints:

[0100] In the formula, R w represents the wheel radius, d represents the wheel track, F x represents the total longitudinal force, and constraint (27) represents that the total driving force distributed to the four wheels needs to meet the requirements of the speed controller. Constraint (28) represents that the yaw moment formed by the longitudinal tire force needs to meet the calculation result of the DYC agent. Constraint (29) represents that the size of the driving force should be within the range of the actuator. T min is the minimum driving torque (i.e. braking torque), T max is the maximum driving torque, and T max = -T min . Equation (26) is solved in MATLAB using the fmincon command, and the four-wheel driving / braking torque is obtained.

[0101] Step eight: send the calculation results of the steering agent and the DYC agent to the actuator.

[0102] Based on the above control method, the embodiment of the present application further proposes:

[0103] An automatic driving vehicle controller, which can execute the content of the above control method.

[0104] A storage medium, which stores the program code of the above control method.

[0105] The above series of detailed descriptions are only specific descriptions for the feasible embodiments of the present application, and are not used to limit the protection scope of the present application. Any equivalent means or changes without departing from the technology of the present application shall be included in the protection scope of the present application.

Claims

1. A cooperative control method for a multi-agent system chassis of an autonomous vehicle based on Co-DMPC, characterized in that, Including the following: S1: Establish a two-degree-of-freedom vehicle model with four-wheel steering and direct yaw moment control (DYC); S2: Establish a distributed state-space equation; based on the different control variables, divide the centralized state-space equation of the vehicle into two agents: a steering agent and a DYC agent. For the steering agent, the control variable of the state-space equation is the front wheel steering angle δ. f and rear wheel steering angle δ r The state variable is the lateral velocity v. y With respect to the vehicle's lateral displacement Y; for the DYC agent, the control variable in the state-space equation is the direct yaw moment M. z The tracked state variables are the yaw rate γ and the yaw angle. S3: Construct a collaborative control framework; including: The input layer uses the vehicle model to calculate the reference yaw rate γ based on the deviation between the vehicle's current state and the target path. ref and reference lateral velocity v yref ; The collaborative control layer decomposes the centralized chassis control system into a distributed control system composed of multiple controllers, enabling each controller to track its own desired target. Through the state-space equation established by S2, the Co-DMPC algorithm is adopted to select a cost function with a global performance objective. Within a unit sampling time, an iterative update method is used, with each agent independently calculating its local optimization and exchanging calculation results with each other multiple times, ultimately enabling the entire system to obtain the global optimal solution. The torque distribution layer optimizes the distribution of driving or braking torque among the four wheels based on the total driving force calculated by the PID speed controller and the direct yaw torque calculated by the collaborative control layer. The execution layer applies control signals to the four wheels via steering actuators and hub motors, enabling precise control of the vehicle. S4: Establish an information transmission protocol; use the same prediction time domain N in all agents. p With control time domain N c For the i-th agent, in the control time domain [t, t+N] c Within [the context], the control quantity predicted at time t+k is denoted as u. i (k|t), define the following three sets of control trajectories: Predicted control trajectory; Optimal control trajectory; Assume a control trajectory; in, This means predicting the control quantity for future moments based on the current state and the control quantity from the previous moment; This represents the optimal solution obtained after solving a locally optimal problem; This indicates that agent i passes information to neighboring agents. The assumed control quantity of the body is obtained by shifting the phase of the predictive control quantity by one sampling period. In the prediction time domain [t, t+N p Within [the context], the state variable predicted at time t+k is denoted as x. i (k|t), define the following three sets of state trajectories: Predict the state trajectory; Optimal state trajectory; Assume a state trajectory; Agent i uses the predicted state of future times at time t. Hypothetical states from neighboring agents With assumption control Solve the distributed optimization problem when k=0. Where, x i (t) and x j (t) represents the state of the vehicle that can be directly measured at the current moment, after obtaining the optimal solution state. With optimal solution control Then, control the first trajectory Applying this to actual control, the optimal state trajectory and the optimal control trajectory are recursively extrapolated step by step to obtain the hypothetical trajectory, i.e., let: At this point, the length of the control trajectory is N. c -1, to complete the last item, let The vehicle's state at the next moment is as follows:

2. The cooperative control method for a multi-agent system of an autonomous vehicle chassis based on Co-DMPC according to claim 1, characterized in that, The two-degree-of-freedom model of the vehicle in S1 is as follows: In the formula, v x Let v be the longitudinal velocity. y γ is the lateral velocity, m is the yaw rate. s For vehicle quality, C αf and C αr These are the front and rear wheel lateral stiffness, l f and l r δ represents the distance from the center of the front and rear wheels to the center of mass, respectively. f and δ r These are the front and rear wheel steering angles, I z M is the moment of inertia of the vehicle rotating about the z-axis. z This is the direct yaw moment; Combining path tracking, the following dynamic equations can be obtained: In the formula, Y represents the lateral displacement of the vehicle. It is the lateral angle.

3. The cooperative control method for a multi-agent system of an autonomous vehicle chassis based on Co-DMPC according to claim 2, characterized in that, In S2, the state-space equation of the steering agent is: In the formula, 4. A cooperative control method for a multi-agent system of an autonomous vehicle chassis based on Co-DMPC, as described in claim 3, is characterized in that... In S2, the state-space equation of the DYC agent is: The state-space equations are: In the formula, 5. A cooperative control method for a multi-agent system of an autonomous vehicle chassis based on Co-DMPC, as described in claim 4, is characterized in that... In the cooperative control layer, the Co-DMPC algorithm includes: S5: Establish the local agent's state equations; including: S5.1 The local continuous linear state equations for the multi-agent system are established as follows: In the formula, m is the number of agents; S5.2 Discretizes the above equation as follows: In the formula: T represents the sampling period; S5.3 is obtained through recursion. Within the time domain The expression: S5.4 The above recursive process can be written in the form of an augmented matrix as follows: In the formula:

6. A cooperative control method for a multi-agent system of an autonomous vehicle chassis based on Co-DMPC, as described in claim 5, is characterized in that... In the cooperative control layer, the Co-DMPC algorithm further includes: S6: Solving local optimization problems; specifically including: S6.1 The local optimization problem considering global performance metrics is expressed as: The equation is subject to the following constraints: In the formula, U ii λ is the set of constraints that control the input. i , λ j Let be the coupling coefficient of the cost function between different agents, and w i (t+k) is the state reference value of agent i, W i For w i The reference state sequence consisting of (t+k), Q i With R i These are the state weight coefficient and the control weight coefficient, respectively. Specifically: Q i It is a penalty for the deviation between the actual trajectory and the expected trajectory, reflecting the vehicle's expectation to follow its desired trajectory; R i This is a penalty for the control input, reflecting the vehicle's expectation to drive with lower control energy. Constraint (17) reflects the vehicle's dynamic constraints, while constraints (18)-(19) require the vehicle's prediction time domain... The initial state must be equal to the actual state of the vehicle. Constraint (20) is the input constraint of the agent to ensure that the optimization result does not exceed the action range of the actuator. S6.2 To improve the controller's computational efficiency, the optimization problem is transformed into a quadratic programming problem: In the formula, const is a constant term that does not affect the optimization result, and S6.3 completes a single optimization calculation, and the local agent obtains a local optimal solution.

7. A cooperative control method for a multi-agent system of an autonomous vehicle chassis based on Co-DMPC, as described in claim 6, is characterized in that... In the cooperative control layer, the Co-DMPC algorithm further includes: A cyclic iterative update approach is adopted, in which each agent exchanges the optimal control sequence and optimal state sequence of local optimization multiple times within a unit sampling time to achieve the solution of the global optimum; for the iterative update, two termination conditions are designed, specifically including: First, let the error between two consecutive iterations of optimization be denoted as . This represents the optimization result of agent i in the nth iteration, where the iteration error Err is... i Less than the threshold ε i When i=1 represents the iteration error of the steering agent and i=2 represents the iteration error of the DYC agent, it is considered that the controller optimization result has converged to the global optimum, and the iteration ends at this time; if the iteration error cannot converge to ε i Then set the maximum number of iterations p. max When the number of iterations is greater than p max The calculation will terminate when the time is reached.

8. A cooperative control method for a multi-agent system of an autonomous vehicle chassis based on Co-DMPC, as described in claim 7, is characterized in that... The specific implementation of the torque distribution layer includes the following: First, the overall driving force equation of the vehicle is constructed based on the PID algorithm: In the formula, K p K represents the proportionality coefficient. i K represents the integral coefficient. d Representing the differential coefficient, e(t) = v x -v xref ; Secondly, consider the following optimization problem. J represents the abbreviation for tire force distribution optimization problem, F xij F represents the longitudinal force on the left front wheel, right front wheel, left rear wheel, and right rear wheel. zij F represents the vertical force on the left front wheel, right front wheel, left rear wheel, and right rear wheel. yij Representing the lateral forces of the four tires, ij = fl, fr, rl, rr represent the left front wheel, right front wheel, left rear wheel, and right rear wheel, respectively. The steering agent has calculated the steering angles of the four wheels, which means that the lateral force F of the tires... yij Since is a constant, it has no effect on solving the above optimization problem. Therefore, formula (25) can be simplified to The above formula is subject to: In the formula, R w The wheel radius is represented by constraint (27), which means that the total driving force distributed to the four wheels must meet the requirements of the speed controller; constraint (28) means that the yaw moment formed by the longitudinal tire force must meet the calculation results of the DYC agent; and constraint (29) means that the magnitude of the driving force should be within the range of the actuator. min The minimum driving torque, i.e., the braking torque, is T. max For the maximum driving torque, and T max =-T min Solving formula (26) yields the four-wheel drive / braking torque.

9. An autonomous vehicle controller, characterized in that, The controller is capable of performing any one of claims 1-8.

10. A storage medium, characterized in that, The storage medium stores the program code of the control method according to any one of claims 1-8.

Images