Adaptive Multi-Robot Coordination: Integrating Improved Potential Fields with Multi-Agent Reinforcement Learning

As the cost of mobile robots decreases, employing multiple robots for complex tasks to enhance efficiency of implementation becomes increasingly viable. Coordinating robots to achieve multiple targets in dynamic environments with limited local information is challenging. Multi-agent reinforcement learning demonstrates much promise in enhancing robot collaboration, yet its effectiveness in partially observable environments remains a challenge. This study proposes a novel multi-agent reinforcement learning framework incorporating artificial potential field information to address this issue. We present an improved artificial potential field method for extracting environmental information and integrate it into our multi-agent reinforcement learning framework, enabling cooperative path planning among agents. Simulations and real-world experiments on robotic platforms demonstrate the efficacy of our approach in improving multi-agent coordination and task performance in complex environments. Our work contributes to the advancement of multi-agent reinforcement learning algorithms for practical robotic applications, offering insights into combining classical control methods with modern learning-based techniques.