Fast Exploration and Coverage of Unknown Indoor Environments Using Real-Time Semantic SLAM on Quadruped Robots

Abstract

Efficient autonomous search and full coverage are essential for quadruped robots in indoor inspection and search-oriented tasks. We present a real-time semantic simultaneous localization and mapping framework that runs entirely onboard a quadruped robot to enable fast indoor exploration. Our system fuses light detection and ranging (LiDAR) sensor and first-person-view (FPV) camera sensing with lightweight segmentation models for semantic classification and traversable-ground detection, and builds structured maps with room, corridor, wall, and obstacle attributes for navigation and exploration. A semantic-guided strategy that combines frontier-based exploration with sampling-based planning reduces path redundancy and improves efficiency in cluttered indoor scenes. Comparative evaluations and real-robot experiments in office-like environments show that the framework achieves a favorable tradeoff between perception accuracy and onboard real-time performance, providing reliable autonomous coverage for complex indoor environments.