EPO-ImDDPG: Evolutionary Policy Optimization Approach with Improved DDPG for Multi-Agent Exploration in Dynamic Environment

Effective multi-agent navigation and path planning in dynamic environments are vital for deploying robotic systems in real-world applications, including search and rescue, surveillance, and industrial operations. This paper presents EPO-ImDDPG (Evolutionary Policy Optimization with Improved Deep Deterministic Policy Gradient), a robust and scalable framework for decentralized multi-agent path planning. The method combines evolutionary policy optimization with reinforcement learning, enabling enhanced training efficiency, seamless scalability, and faster convergence in challenging and dynamic scenarios. The proposed cooperative exploration approach utilizes dynamic Voronoi partitions to minimize duplicated exploration areas, effectively assigning different target locations to individual robots. This ensures efficient coverage of the environment and reduces overall exploration redundancy. Experiments demonstrate that EPO-ImDDPG achieves 25% higher success rates and more stable performance when exposed to a large number of dynamic obstacles, outperforming traditional reaction-based planners like LRA*, as well as state-of-the-art learning-based methods such as DDPG and TD3.