セマンティックセグメンテーションを用いた走行可能領域判定に基づく人追従ロボット
Abstract
This study proposes a human-following robot system designed to assist the elderly with daily mobility and carrying loads in rural and mountainous areas. To ensure safe navigation in such unstructured environments, our approach combines semantic segmentation with object detection. Using an RGB-D camera, the system applies semantic segmentation to classify traversable and non-traversable areas, such as grass, generating a 2D costmap. Simultaneously, the target person is tracked using YOLO11. An optimal path avoiding hazardous regions is then generated using the A* algorithm and executed via the Pure Pursuit method. Field experiments conducted with a mobile robot demonstrated that the proposed method achieved a 100% following success rate and improved the grass area avoidance rate from 0% to 80% compared to conventional methods. The results confirm the system’s effectiveness in autonomously avoiding difficult terrain while safely tracking a human.
