Research On Adaptive Learning Control Of Multi-agent Cooperative Formation

In recent years,multi-agent systems have attracted more and more attention from scholars at home and abroad.Compared with single-agent,multi-agents can collaboratively complete single-and multi-objective tasks that are difficult for single-agent to complete.Multi-agent coordinate control technology is still one of the key research directions today.Among them,consensus control is the basic problem of multi-agent coordinate control.Later,coordinate tracking,formation,clustering,and containment control problems have been developed based on consensus control.Considering the existence of multiple collision avoidance and formation control requirements for multiagent system,it is necessary to design different controllers for collision avoidance and formation control problems under different control requirements.Based on actor-critic learning method,this thesis studies the multi-agent cooperative control problem under multiple collision avoidance and formation control scenarios,and the main research contents are as follows:First,the solution to the first-order multi-agent consensus control problem is extended to the second-order multi-agent cooperative formation control.The optimal cooperative formation control policy based on actor-critic learning method is implemented,which is solved by generalized policy iteration algorithm and constructed by neural networks.Then,a hybrid learning formation controller is constructed based on artificial potential field method,which effectively avoids collisions among agents.Finally,the simulation verifies the effectiveness of the proposed controller in formation control target and collision avoidance design.Secondly,combined with artificial potential field method and reinforcement learning technology,an anti-collision formation cooperative controller is designed based on actor-critic learning method with the integrated design of collision avoidance and formation control.It is solved by generalized policy iteration algorithm and constructed by neural networks.Through the discussion on the relationship of fixed topologies and switching topologies,the controller can be implemented under switching topologies.Finally,the simulation verifies the effectiveness of the proposed algorithm in anti-collision formation cooperative control.Finally,for large-scale multi-agent systems,a hierarchical leader-following control structure based on multi-group technology is proposed,and a formation controller based on actor-critic learning method is designed.Based on generalized policy iteration algorithm,multi-step generalized policy iteration algorithm is designed with the improvement of policy evaluation.It can further accelerate the convergence speed and reduce the running time.Finally,the simulation verifies the performance of the hierarchical control structure and the effectiveness of the proposed algorithm in cooperative formation control.