Optimized Deep Reinforcement Learning for Cooperative Path Planning and Obstacle Avoidance in AGV Dynamic Environments

Abstract

As smart logistics and factories advance, Automated Guided Vehicles (AGVs) face increasingly complex and dynamic environments, requiring more intelligent path planning and obstacle avoidance. Traditional path planning algorithms, while widely used, often suffer from high computational costs and limited generalization capabilities. This paper introduces an optimized approach utilizing Deep Deterministic Policy Gradient (DDPG) and Multi-Agent DDPG (MADDPG) algorithms to address AGV cooperative path planning in dynamic scenarios. By modeling AGVs as agents within a deep reinforcement learning framework, a centralized training and decentralized execution approach is employed. The algorithms are enhanced through optimized experience buffer sampling methods, allowing agents to autonomously navigate and avoid static and dynamic obstacles.