Robust formation tracking of multi-agent systems via reinforcement learning-based hybrid impulsive control.
TL;DR
The Hierarchical Multi-agent Cooperative Reinforcement Learning (HMAC-RL) framework is introduced to optimally and adaptively refine both control parameters and impulsive moments and offers enhanced adaptability, faster corrective actions, and improved system robustness in uncertain and evolving environments.
Abstract
This paper explores robust H∞ formation tracking control of multi-agent systems in terms of hybrid impulsive approach combined with reinforcement learning. The study establishes the robust stabilization of the formation under the proposed control protocol by employing the Razumikhin technique and implementing feasible constraints to ensure sustained robust H∞ performance. Compared to recent continuous and impulsive control methods, the developed hybrid impulsive control framework offers enhanced adaptability, faster corrective actions, and improved system robustness in uncertain and evolving environments. Furthermore, in contrast to the majority of current formation stabilization techniques, we introduce the Hierarchical Multi-agent Cooperative Reinforcement Learning (HMAC-RL) framework to optimally and adaptively refine both control parameters and impulsive moments. This framework features a central agent module, a continuous control agents module, and a pulsed control agents module, each serving crucial roles in addressing specific aspects of the hybrid impulsive control protocol design. Finally, numerical simulations are presented to support the theoretical analysis and demonstrate the optimization efficiency achieved through the reinforcement learning framework HMAC-RL.