Obstacle avoidance inspection method of cable tunnel for quadruped robot based on particle swarm algorithm and neural network

• Published in: Scientific reports Vol. 15; no. 1; pp. 36065 - 18
• Main Authors: Wu, Jianfeng, Huang, Yaosheng, Lai, Yingdong, Yang, Shangbin, Zhang, Chao
• Format: Journal Article
• Language: English
• Published: London Nature Publishing Group UK 15.10.2025; Nature Publishing Group; Nature Portfolio
• Subjects: 639/166; 639/766/259; Accuracy; Algorithms; Cable tunnel; Convolutional neural network; Design; Efficiency; Fuzzy logic; Hip joint; Humanities and Social Sciences; Inspection; Knee; Legs; multidisciplinary; Neural networks; Obstacle avoidance inspection; Optimization algorithms; Parameter estimation; Particle swarm optimization algorithm; Quadrupedal robot; Robots; Science; Science (multidisciplinary); Sensors; Walking; Cable tunnel; Obstacle avoidance inspection; Convolutional neural network; Quadrupedal robot; Particle swarm optimization algorithm
• ISSN: 2045-2322; 2045-2322
• DOI: 10.1038/s41598-025-19903-w

The cable tunnel environment is characterized by its narrow confines and the presence of numerous irregularly placed obstacles such as pipes and supports, which pose significant challenges to inspection operations. To address this issue, this paper proposes an obstacle avoidance inspection method for quadruped robot cable tunnel based on particle swarm optimization and neural network. Through an analysis of the supporting and swinging phases of the quadruped robot’s legs when walking, its dynamic performance is calculated to determine the inaccessible areas within the tunnel. Subsequently, a map is constructed that delineates known areas, restricted zones, unknown regions, and obstacles. A VGG-16 neural network is employed to detect and localize the obstacles in the robot’s path, while a particle swarm optimization algorithm is utilized to plan the optimal inspection route. The algorithm demonstrates rapid convergence, requiring only 11 iterations and 1.335 s to complete, with a stable training error of approximately 0.135. Its performance is deemed satisfactory, offering an effective solution for the efficient inspection of cable tunnels by quadruped robots.