Integrated Robotic System for Autonomous Inspection of Power Transmission Lines Using Multimodal Perception and Reconfigurable Topologies

This paper presents an integrated robotic system for autonomous power transmission line inspection, evaluating three mechanical topologies—Line Walker, ModuClimber, and FlexRover—and a multimodal perception suite combining depth sensors. The comparative analysis reveals a distinct trade-off between mechanical simplicity and obstacle negotiation capabilities, with complex articulated designs proving essential for traversing obstacles. Crucially, the study uncovers significant challenges in algorithm portability; machine learning models trained on a specific robot topology suffered severe performance degradation when transferred to others due to geometric sensor variations. However, results demonstrate that fusing geometric depth features with depth statistics allows lightweight classifiers to recover up to 100% accuracy across different platforms. The findings establish that while deep learning models like SqueezeNet offer inherent robustness, feature-based sensor fusion is the key enabler for developing portable autonomous inspection systems. Categories: (3), (4), (8)