A Robust Human-Following System for Autonomous Mobile Robot in Unknown Environments

Abstract

Autonomous human-following remains one of the core challenges in mobile robotics and human-robot collaboration, playing a pivotal role in enhancing robotic intelligence. This paper presents a two-stage system integrating sensor fusion and trajectory optimization to address human-following challenges in unknown environments. The first stage introduces a robust person tracking method: leveraging LiDAR detection data with a motion filtering mechanism, we improve detection accuracy through camera association, while Ultra-Wideband devices mitigate target ambiguity in multi-person scenarios and fuse with LiDAR data to generate smooth and continuous trajectories. Crucially, the system maintains operational reliability in complete darkness through graceful degradation to a Camera-Free Mode. The second stage employs a local dynamic map to search for collision-free and dynamically feasible trajectories, enhanced through B-spline optimization for improved smoothness and safety. We introduce a novel adaptive strategy that dynamically adjusts trajectory temporal parameters based on target distance to ensure stable tracking. The system can be deployed on resource-constrained platforms. Experimental results show that the proposed human-following system can accurately identify target individuals among multiple pedestrians, handle occlusions effectively, and follow the target robustly while avoiding obstacles in unknown environments. The complete experimental video is available at https://youtu.be/6a3i7ua_14k.