Reinforcement Learning Control Design for Multi-Agent Robots: A Model-Free Approach

This article delves into the development of a reinforcement learning (RL) controller tailored for multi-agent robots, specifically focusing on a system with two interacting agents. The selected model-free RL controller is designed to adeptly handle uncertainties and unknown parameters within the complex dynamics of multi-agent setups. The study employs Q Learning for agent training, aiming not only to foster consensus among agents but also to minimize tracking errors for individual robots. The introduced central control model refines the mathematical foundations of RL, with a focus on optimizing Consensus Tracking as a regulatory mechanism. The article evaluates the controller's performance through simulations, emphasizing its effectiveness in managing intricate interactions between the agents.