# Two key algorithms for intelligent inspection robots in electric bicycle charging sheds

## TL;DR

This paper introduces two new algorithms to improve the performance of inspection robots in narrow electric bicycle charging sheds, enhancing safety and efficiency.

## Contribution

The paper introduces MS-RRT* for narrow path planning and an improved SOLOv2 for small-target recognition in inspection robots.

## Key findings

- MS-RRT* reduces sampling nodes in obstacles by 29.36% and achieves 100% exploration success in narrow channels.
- Improved SOLOv2 increases small-target detection accuracy from 52.9% to 62.5% without slowing processing speed.
- The algorithms outperform existing methods in efficiency and robustness for inspection tasks.

## Abstract

The deployment of intelligent inspection robots in electric bicycle charging sheds is critical for preventing fire hazards, yet faces challenges in navigating narrow passages and recognizing small components. This paper proposes two enhanced algorithms to address these issues: (1) a multi-root node RRT* (MS-RRT*) for efficient narrow-channel path planning, and (2) an improved SOLOv2-based instance segmentation method for small-target recognition. The MS-RRT* introduces dynamic secondary root nodes with constrained expansion cycles, significantly increasing the probability of traversing narrow channels while reducing sampling nodes in obstacles by 29.36% compared to classical RRT*. For component recognition, the enhanced SOLOv2 algorithm augments feature pyramid outputs with larger hierarchical maps, improving small-target accuracy (e.g., button detection from 52.9% to 62.5%) without compromising processing speed. Experimental results demonstrate that the proposed MS-RRT* achieves a 100% exploration success rate in narrow channels, outperforming state-of-the-art methods in both efficiency and robustness. The improved SOLOv2 also surpasses Mask R-CNN in multi-category component recognition, ensuring reliable inspection in complex scenarios. These advancements collectively enable 24/7 automated monitoring, addressing critical safety demands in real-world charging infrastructure.