Gesture-Based Human-Robot Interaction Framework for Teleoperation Control of Agricultural Robot

Abstract

Remote control based on data-driven has been widely studied in human-robot interaction, which converts human body movements into digital signals through somatosensory devices for teleoperation control. It is a convenient and efficient intelligent solution to handle the robot to realize the agricultural task in a complex environment. In this paper, a real-time dynamic recognition framework based on a three-dimensional convolutional neural network is designed for agricultural robot. The deep separable convolution and inverted residual structure are presented, and then a gesture classification network architecture is discussed. Finally, some demonstrations using Unitree agricultural robot are carried out to illustrate the hand gesture application with an efficient response.