Reinforcement Learning-Based Fixed-Time Consensus for Multiple Uncertain Euler-Lagrange Systems with Variable Connectivity Preservation
Ping Li, Ли Фу, Zhibao Song · 2025
This note works on building a novel actor-critic reinforcement learning(RL) based controller to implement the optimized fixed-time consensus for uncertain leader-following Euler-Lagrange multi-agent systems. Compared with the previous results, the communication between two agents is limited by variable distances. By introducing a class of dynamic bounded potential functions, the connectivity of topology graph is preserved. Then, the consensus issue is transformed into a novel multi-layer tracking control scheme by refining the relationship between parent nodes and its child nodes in a directed minimum spanning topology, included in the initial topology graph. With the present method, the controllers for all child nodes are hierarchically constructed to track its parent node, and finally achieve consensus with the leader. Moreover, the convergent rate of critic neural network (NN) is improved by the aid of a fresh fixed-time gradient algorithm, ensuring the validity of guidance for critic NN. Simulation example is the final logical step to present a comparison with the traditional fixed-time NN controller, which illustrates the superiority of the invoked control algorithms.