Zhiming Fan, Kefeng Li, Guangyuan Zhang, Zhenfang Zhu, Peng Wang, Zhenfei Wang, Chen Fu, Xiaotong Li, Yongshuo Zhang, Yongpeng Zhao
DWA, DDPG, local path planning, reinforcement learning
In recent years, mobile robots have been widely applied across various fields, where local path planning technology plays a crucial role in real-time obstacle avoidance and trajectory adjustment. The dynamic window approach (DWA), as a classical local path planning algorithm, still exhibits significant limitations in practical applications: its fixed-weight evaluation function results in poor environmental adaptability, and in obstacle-dense areas, it is prone to collisions or abrupt braking failures due to insufficient turning space. To address these issues, this study proposes an adaptive enhanced DWA algorithm based on deep deterministic policy gradient (DDPG). First, the DWA evaluation function is optimised to incorporate obstacle density prediction, thereby reserving safe avoidance space. Second, a continuous action– state space DDPG model is constructed to dynamically adjust the evaluation function weights through reinforcement learning. Experimental results demonstrate that this method significantly enhances the planning success rate and environmental adaptability of the DWA algorithm in complex scenarios, providing an effective solution to the inflexibility of traditional approaches.
Important Links:
Go Back
