Local Semantic Map-Based Navigation System for Quadruped Inspection Robot in Complex Terrains

Traditional navigation methods provide a convenient and effective approach for quadruped robot navigation. However, the lack of terrain height awareness often leads to misidentifying height variations as obstacles, limiting their effectiveness in inspection tasks. This paper proposes a navigation system for quadruped inspection robots based on local semantic map, utilizing a convolutional neural network-based semantic segmentation algorithm to obtain real-time environmental information. By constructing a multi-level global costmap and a local semantic map through a hierarchical costmap framework, the proposed method enables robust navigation in complex terrains and dynamic obstacle environments. This article was validated in simulated and actual environments including slopes, steps, etc., with a success rate of 85.7%. This approach enhances the adaptability of quadruped robots, addressing navigation challenges while improving inspection efficiency and safety.