Continuous Dynamic Obstacle Environment Path Optimization Algorithm for continuous dynamic obstacle environment

Abstract This paper delves into the challenges of continuous path planning for robots in complex environments, addressing the limitations of traditional algorithms in handling sparse rewards and insufficient exploration.To tackle these issues, we propose an improved Deep Deterministic Policy Gradient (DDPG) algorithm named DDAPG, which integrates a potential energy-based reward reshaping method and an adaptive round Ornstein-Uhlenbeck (OU) noise model.The reward reshaping method enhances the algorithm's ability to provide smooth and continuous rewards, guiding the robot towards the target while avoiding obstacles effectively.Meanwhile, the adaptive OU noise model balances exploration and exploitation by adjusting the noise level according to the training progress, preventing the robot from getting stuck in local optima.Experimental results in a continuous simulation environment demonstrate that the DDAPG algorithm significantly improves convergence speed, reduces collision rates, and enhances adaptability compared to baseline algorithms.Specifically, DDAPG achieves a success rate of over 92% and a low collision rate of 7.625% in reaching target positions while navigating around obstacles.These findings underscore the potential of DDAPG in advancing embodied intelligence for robot path planning, offering a robust framework for deploying robots in real-world scenarios with dynamic and complex environments.