<p>In modern industrial settings, effective human-robot interaction (HRI) is critical for enabling safe, intuitive, and efficient collaboration between human operators and mobile robots. While gesture-based control has been widely adopted, most existing systems focus primarily on robotic arms, offering limited adaptability in mobile robot navigation. In particular, current mobile robot systems often lack the capability for operators to dynamically influence or modify the robot’s path in real time. To address this limitation, we propose a novel HRI framework that allows a mobile robot to dynamically adjust its navigation trajectory based on the contextual interpretation of operator gestures and activities. Unlike prior approaches that are restricted to static robot interaction, our system enables responsive, on-the-fly adaptation of movement during live operations. To support this functionality, we introduce a custom RGB video dataset specifically designed for industrial navigation contexts and train a Temporal Shift Module (TSM)- based model capable of recognizing both simple gestures and broader operator activities. The system is integrated into the ROS 2 Nav2 navigation stack, and we develop a custom plugin that modifies the robot’s trajectory in real time based on recognized commands—such as bypassing the operator on a specified side or stopping and resuming movement. Experimental evaluations in a semi-realistic industrial environment show promising results.</p>