A Visual Servo Human-Robot Interaction Method Based on Unidirectional Adaptive Temporal Interaction

Gesture recognition holds a pivotal role within the realm of human-robot interaction. Existing methods for action recognition often exhibit suboptimal real-time performance and face challenges in deployment onto edge devices for instantaneous online recognition. Moreover, human-robot interaction typically relies on wearable devices, with limited purely vision-centric interaction methodologies available. This study introduces the Unidirectional Adaptive Temporal Interaction Module (Unidirectional-ATIM), which integrates unidirectional feature fusion across temporal dimensions to capture historical frame-based temporal information. Drawing from this module, we developed a lightweight online action recognition algorithm, ATIM-Online, and established a visual servo-based framework for gesture human-robot interaction. Experimental out comes from deployments on quadruped robots and robotic arms demonstrate that our approach achieves high accuracy in gesture recognition and a high success rate in interaction.