Autonomous Adjustment of Tracking Position in Dynamic Environments for Human-Following Robots Using Deep Reinforcement Learning

Abstract

Achieving flexible human-following in real-world environments remains a critical yet challenging problem in Human-Robot Interaction (HRI). Traditional approaches typically constrain robots to fixed tracking positions—such as following from behind, ahead, or alongside—thereby limiting their adaptability in dynamic and unstructured environments. This study introduces a reinforcement learning-based framework that allows the robot to dynamically adjust its tracking positions in response to workspace constraints. An interaction space is defined to capture the relationship between the human and the robot while considering the environment. This space serves as the basis for state spaces in Deep Reinforcement Learning (DRL), helping the robot adapt to environmental changes. The selected tracking position is then utilized as input for a human-following controller, ensuring smooth and continuous motion. Experimental evaluations in both indoor and outdoor environments demonstrate that the proposed approach enables robots to follow humans flexibly and adaptively, adjusting their positions autonomously and avoiding obstacles without requiring a predefined tracking position.