Deep Q‐Network‐Based Hierarchical Path Planning of a Biped Wall‐Climbing Robot for Window Cleaning

Abstract

In building maintenance, cleaning the exterior walls of high‐altitude buildings is a critical task traditionally performed by human workers. However, manual maintenance exhibits several challenges, including fall risks, high costs, and aging workforce. To address such issues, this article presents a biped wall‐cleaning robotic system (BWCRS) dedicated to the maintenance of multi‐isolated areas on exoskeleton‐structured windows. To solve the complete coverage cleaning path planning problem, a full coverage cleaning approach is proposed by integrating a deep Q‐network (DQN)‐based method with the BWCRS. The proposed method enhances the cleaning process by optimizing the robot's travel path using DQN, ensuring stable navigation on walls with obstacles through advanced biped loco‐manipulation. The effectiveness of the reported approach is validated through extensive simulations and experimental investigations. Comparative studies reveal that the presented method outperforms the existing method. The results demonstrate the BWCRS's capability to efficiently and safely clean exoskeleton‐structure windows. Moreover, the proposed framework can be extended to path planning of other types of climbing robots.