Hand Gesture-based Teleoperation Control of a Mecanum-wheeled Mobile Robot

Abstract

The aim of this study is to design an interface for the teleoperation control of a Mecanum-wheeled Mobile Robot (MMR). To make a cost-effective and non-invasive teleoperation interface approach, a hand gesture recognition strategy via a webcam is adopted. In this study, seven hand gestures are recognized on a desktop PC and then the corresponding velocity commands are sent to the MMR via WiFi. The SqueezeNet architecture is employed for the hand gesture recognition and MMR teleoperation task after various Convolutional Neural Network (CNN) models are compared. The comparison considers some aspects, i.e., accuracy, memory usage, and recognition speed, to find the most feasible method for real-time hand gesture recognition and teleoperation control. Experimental results are presented to show the excellence of the proposed SqueezeNet architecture.