Title: GCN-LSTM-based robot patrol path planning model for unknown environments

Authors: Xiaohan Wang

Addresses: School of Electromechanical and Automotive Engineering, Jiuzhou Polytechnic, Xuzhou, 221116, China

Abstract: Robot inspection in unknown dynamic environments faces challenges in incomplete perception and limited real-time obstacle avoidance. Its novelty lies in its ability to capture the temporal motion characteristics of dynamic obstacles through modelling. It integrates a multi-objective reward mechanism with rolling time-domain optimisation to achieve dynamic feature fusion and incremental risk prediction. This approach enhances both the global efficiency and local responsiveness of path planning. Experiments showed local/global planning times of 0.10 s/1.78 s, 66.7%-78.7%, and 64.5%-68.7% lower than DDPG, SAC, GAT, and MTR. Mean path deviation was 0.19 m, dynamic obstacle detection rate 91.36%, and response latency 32.19 ms, outperforming benchmarks. The research's significance lies in its proposal of an efficient and safe dynamic path planning model, as well as its substantial enhancement of the system's real-time performance, robustness, and scalability in complex dynamic scenarios through a hierarchical architecture design. This provides a novel collaborative decision-making paradigm for robotic autonomous patrol tasks. [Submitted 13 August 2025; Accepted 6 April 2026]

Keywords: path planning; GCN-LSTM; PPO; MPC; mobile robot.

DOI: 10.1504/IJMR.2026.154550

International Journal of Manufacturing Research, 2026 Vol.20 No.2, pp.231 - 254

Received: 13 Aug 2025
Accepted: 06 Apr 2026
Published online: 03 Jul 2026 *

Full-text access for editors Full-text access for subscribers Purchase this article Comment on this article