Two-stage coupling fusible multi-agent control framework
Through the dual-stage coupling, the multi-agent control framework that can be integrated, combined with the upper-level planning and the lower-level control system, the problem of traditional angle module vehicles lacking efficient control architecture in multi-vehicle collaborative control is solved, and the comprehensive optimization from macro to micro is achieved Control enhances coordination between vehicles.
Patent Information
- Application Number
- CN202410769679.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-14
- Publication Date
- 2024-11-12
Smart Images

Figure CN118945200A_ABST
Abstract
Description
Technical Field
[0001] The present invention belongs to the field of vehicle coordinated control, and in particular relates to a two-stage coupled fusionable multi-agent control framework. Background Art
[0002] In recent years, in order to simplify vehicle design and assembly while reducing costs, the existing technology has designed a modular integrated corner module based on the geometric boundaries of the wheel, integrating the suspension, steering, braking and power systems into one corner module, especially in combination with x-by-wire technology, which can greatly improve the chassis operability. The result of replacing traditional wheels with corner modules is that the wheels of the vehicle are completely decoupled, and the increase in degrees of freedom leads to a surge in the difficulty of coordinated control of each wheel. In recent years, multi-agents have been widely used in the vehicle field, especially in the field of multi-vehicle collaborative control.
[0003] Corner module vehicles lack an efficient control architecture in the multi-vehicle collaborative control process, the upper-level control objectives do not consider the underlying constraints, and the output control objectives conflict with the lower-level execution capabilities. Summary of the invention
[0004] The purpose of the present invention is to provide a two-stage coupled and fused multi-agent control system to solve the problem of the lack of an efficient control architecture in the multi-vehicle collaborative control process of traditional corner module vehicles.
[0005] The technical solution to achieve the purpose of the present invention is: a two-level coupled fusionable multi-agent control framework, the multi-agent control framework includes an upper-level planning system and a lower-level control system;
[0006] The upper-level planning system includes a fusion layer and an upper-level vehicle agent control system. The fusion layer is used to integrate road information, infrastructure information, obstacle information, and multi-vehicle information. The upper-level vehicle agent control system takes the formation and distance between multiple vehicles as planning targets, uses the current state and limit capacity of each vehicle as constraints, forms vehicle-to-vehicle information interaction through V2V communication, and performs multi-vehicle comprehensive Pareto optimal target planning.
[0007] The lower-level control system includes an intelligent agent comprehensive corner module parameter analysis system and a lower-level corner module intelligent agent control system. The intelligent agent comprehensive corner module parameter analysis system takes passability and handling stability as its goals, and adopts a Pareto optimal calculation method to solve multi-objective optimization problems. The lower-level corner module intelligent agent control system is used to establish a non-zero and infinite-duration linear quadratic differential game model in which each intelligent agent of the chassis control system participates, determine their respective payoff functions, construct a linear feedback Nash equilibrium equation, and solve the coupled algebraic Riccati equations through an iterative algorithm to obtain the solution to the Nash equilibrium. According to the solution results, the chassis control intelligent agents exchange and cooperate with each other to dynamically coordinate the driving, braking, steering, and suspension of each corner module intelligent agent.
[0008] Furthermore, the upper-level vehicle agent control system is specifically as follows:
[0009] Agent m (m = 1, 2..) is initialized and generates the initial state u nm (k);
[0010] Agent m (m = 1, 2..) uses the state u at the previous moment nm (k) Calculate the state at time K+1, that is, the agent passes ∑U n ,m(m=1,2.....m) Solve the Pareto optimal function to get U nm (K+1);
[0011] The agent changes its state u at time K+1 nm (k) is sent to other agents, and the agents update the weight W according to the state at time K+1. ω .
[0012] Furthermore, the weight W ω The expression is as follows:
[0013]
[0014] Furthermore, the Pareto optimal function includes single-agent phase plane constraints and phase plane planning.
[0015] Furthermore, the control system of the lower corner module agent is as follows:
[0016] Agent m (m = 1, 2..) is initialized and generates the initial state u nm (k);
[0017] Agent m (m = 1, 2..) uses the state u at the previous moment nm (k) Calculate the state at time K+1, through ∑U n,m(m=1,2.....m) solve the Pareto optimal function;
[0018] The agent changes its state u at time K+1 nm (k) is sent to other agents, and the agent updates the weight W according to the state at time K+1 ω .
[0019] Furthermore, the vehicle agents communicate with each other through the vehicle controller to form a comprehensive optimal solution for the behavior planning of each vehicle agent. Each corner module agent conducts game and comprehensive optimal control based on the comprehensive optimal solution, and finally achieves a comprehensive optimal solution from macro to micro.
[0020] Furthermore, the bottom-level agents of different upper-level agents communicate with each other, and the two upper-level agents are merged, and the corner modules between each vehicle communicate through the wireless network of the upper-level controller.
[0021] Compared with the prior art, the present invention has the following significant advantages:
[0022] The control architecture of the present invention is at the upper layer. Each vehicle agent communicates with each other through the vehicle controller. Through game theory and optimal control theory, a comprehensive optimal solution for the behavior planning of each vehicle agent is formed. The control target of each vehicle agent is input to the lower layer. Each corner module agent plays a game and performs comprehensive optimal control based on the target, and finally achieves a comprehensive optimal solution from macro to micro. At the same time, the bottom agents of different upper-layer agents can also communicate with each other, and the two upper-layer agents are integrated to a certain extent to increase the coordination between the two.
[0023] The present invention regards the corner module as a separate intelligent entity. Each corner module intelligent entity transmits information and performs game control through the controller and vehicle-mounted communication to ultimately achieve the control goal of a single vehicle. In other words, each vehicle itself is an intelligent entity, and there are many corner module intelligent entities at its bottom layer. The two are not on the same level, but influence each other. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] Figure 1 Schematic diagram of the double-layer multi-agent control system of the present invention.
[0025] Figure 2 Schematic diagram of the intelligent agent fusion of the present invention. DETAILED DESCRIPTION
[0026] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the accompanying drawings in the embodiments of the present invention.
[0027] See also Figure 1,The present invention provides a two-level multi-agent control system, including an upper-level planning system and a lower-level control system;
[0028] The upper-level planning system includes a fusion layer and an upper-level vehicle intelligent control system. The fusion layer is used to integrate road information, infrastructure information, obstacle information, and multi-vehicle information. The upper-level vehicle intelligent control system is used to take the formation and distance between multiple vehicles as planning targets, and use the current status and limit capacity of each vehicle as constraints to form vehicle-to-vehicle information interaction through V2V communication, and perform multi-vehicle comprehensive optimal target planning.
[0029] The lower-level control system includes an intelligent agent comprehensive corner module parameter analysis system and a lower-level corner module intelligent agent control system. The intelligent agent comprehensive corner module parameter analysis system takes passability and handling stability as its goals, and plans to use a Pareto optimal calculation method to solve multi-objective optimization problems. The lower-level corner module intelligent agent control system is used to establish a non-zero and infinite-duration linear quadratic differential game model involving each intelligent agent in the chassis control system, determine their respective payoff functions, construct a linear feedback Nash equilibrium equation, and solve the coupled algebraic Riccati equations through an iterative algorithm to obtain the solution to the Nash equilibrium. According to the solution results, the chassis control intelligent agents exchange and cooperate with each other to dynamically coordinate the driving, braking, steering, suspension, etc. of each corner module intelligent agent.
[0030] The upper-level vehicle intelligent control system is:
[0031] Agent m (m = 1, 2..) is initialized to generate the initial state u nm (k);
[0032] Agent m (m = 1, 2..) uses the state u at the previous moment nm (k) Calculate the state at time K+1, that is, the agent passes ∑U n ,m(m=1,2.....m) Solve the Pareto optimal function to get U nm (K+1);
[0033] The agent changes its state u at time K+1 nm (k) is sent to other agents, and the agents update the weight W according to the state at time K+1. ω .
[0034] The control system of the lower corner module agent is:
[0035] Agent m (m = 1, 2..) is initialized to generate the initial state u nm (k);
[0036] Agent m (m = 1, 2..) uses the state u at the previous moment nm(k) Calculate the state at time K+1, through ∑U n ,m(m=1,2.....m) solve the Pareto optimal function;
[0037] The agent changes its state u at time K+1 nm (k) is sent to other agents, and the agents update the weight W according to the state at time K+1. ω .
[0038] The vehicle agents communicate with each other through the vehicle controller to form a comprehensive optimal solution for the behavior planning of each vehicle agent. Each corner module agent conducts game and comprehensive optimal control based on this goal, and finally achieves a comprehensive optimal solution from macro to micro.
[0039] The bottom-level intelligent agents of different upper-level intelligent agents can also communicate with each other, integrating the two upper-level intelligent agents. The corner modules between each vehicle can also communicate through the wireless network of the upper-level controller, so as to achieve the optimal solution under special working conditions.
[0040] When this application is used: the upper layer integrates environmental information such as road information, infrastructure information, obstacle information, and multi-vehicle information, takes the formation and distance between multiple vehicles as planning targets, forms vehicle-to-vehicle information interaction through V2V communication based on constraints such as the current state and limit capacity of each vehicle, and performs multi-vehicle comprehensive optimal target planning; after the planning is completed, the optimal target of each vehicle is sent to the lower-layer multi-agents;
[0041] The lower-level multi-agents comprehensively analyze factors such as the corner module parameter constraints and handling performance constraints, and take passability and handling stability as the goals. It is proposed to use a Pareto optimal calculation method to solve the multi-objective optimization problem. The coupling relationships such as dynamic coupling, tire force coupling, and vertical load coupling in the movement of reconstructed special vehicles can lead to functional overlap and control conflict problems between chassis control agents. The differential game method is used to establish a non-zero and infinite-length linear quadratic differential game model involving each agent of the chassis control system, determine their respective payoff functions, construct a linear feedback Nash equilibrium equation, and solve the coupled algebraic Riccati equations through an iterative algorithm to obtain the solution to the Nash equilibrium. According to the solution results, the chassis control agents exchange information and cooperate with each other to dynamically coordinate the driving, braking, steering, suspension, etc. of each corner module agent.
[0042] In particular, see Figure 2 ,The corner modules between each vehicle can also communicate through the wireless network of the upper controller, forming a certain degree of ,fusion, which can achieve certain special working conditions, such as two vehicles working together to transport a large item.
Claims
1. A two-level coupled and fusionable multi-agent control framework, characterized in that: The multi-agent control framework includes an upper-level planning system and a lower-level control system; The upper-level planning system includes a fusion layer and an upper-level vehicle agent control system. The fusion layer is used to integrate road information, infrastructure information, obstacle information, and multi-vehicle information. The upper-level vehicle agent control system takes the formation and distance between multiple vehicles as planning targets, uses the current state and limit capacity of each vehicle as constraints, forms vehicle-to-vehicle information interaction through V2V communication, and performs multi-vehicle comprehensive Pareto optimal target planning. The lower-level control system includes an intelligent agent comprehensive corner module parameter analysis system and a lower-level corner module intelligent agent control system. The intelligent agent comprehensive corner module parameter analysis system takes passability and handling stability as its goals, and adopts a Pareto optimal calculation method to solve multi-objective optimization problems. The lower-level corner module intelligent agent control system is used to establish a non-zero and infinite-duration linear quadratic differential game model in which each intelligent agent of the chassis control system participates, determine their respective payoff functions, construct a linear feedback Nash equilibrium equation, and solve the coupled algebraic Riccati equations through an iterative algorithm to obtain the solution to the Nash equilibrium. According to the solution results, the chassis control intelligent agents exchange and cooperate with each other to dynamically coordinate the driving, braking, steering, and suspension of each corner module intelligent agent.
2. The multi-agent control framework according to claim 1, characterized in that: The upper-level vehicle agent control system is specifically as follows: Agent m (m = 1, 2..) is initialized and generates the initial state u nm (k); Agent m (m = 1, 2..) uses the state u at the previous moment nm (k) Calculate the state at time K+1, that is, the agent passes ∑U n ,m(m=1,2.....m) Solve the Pareto optimal function to get U nm (K+1); The agent changes its state u at time K+1 nm (k) is sent to other agents, and the agents update the weight W according to the state at time K+1. ω .
3. The multi-agent control framework according to claim 2, characterized in that: Weight W ω The expression is as follows:
4. The multi-agent control framework according to claim 3, characterized in that: The Pareto optimal function includes single-agent phase plane constraints and phase plane planning.
5. The multi-agent control framework according to claim 4, characterized in that: The control system of the lower corner module agent is as follows: Agent m (m = 1, 2..) is initialized and generates the initial state u nm (k); Agent m (m = 1, 2..) uses the state u at the previous moment nm (k) Calculate the state at time K+1, through ∑U n ,m(m=1,2.....m) solve the Pareto optimal function; The agent changes its state u at time K+1 nm (k) is sent to other agents, and the agent updates the weight W according to the state at time K+1 ω .
6. The multi-agent control framework according to claim 5, characterized in that: The vehicle agents communicate with each other through the vehicle controller to form the comprehensive optimal solution for the behavior planning of each vehicle agent. Each corner module agent conducts game and comprehensive optimal control based on the comprehensive optimal solution, and finally achieves the comprehensive optimal from macro to micro.
7. The multi-agent control framework according to claim 6, characterized in that: The bottom-level agents of different upper-level agents communicate with each other, and the two upper-level agents are merged. The corner modules between each vehicle communicate through the wireless network of the upper-level controller.
