GestureCommand: A Simulated Camera-Based Gesture Recognition System for Autonomous Table-Specific Delivery Robot

University of Birmingham
Intelligent Robotics Final Team Project

*Indicates Equal Contribution

Demo video of Ordering robot to go Table 4 and back to till.

Abstract

This document details the development of robotic simulation software, tailored to improve service in cafe environments. Central to this development are two features: gesture control and advanced navigation. The gesture control module enables cafe staff to command the robot to specific tables, and additionally, allows customers to signal the robot to return to the till using straightforward hand gestures. Meanwhile, the navigation feature ensures an accurate and efficient path finding of the robot from the till to the designated table and back. This combination aims to enhance operational efficiency, reduce service duration, and elevate the overall customer experience within cafe settings. This report includes technical parts of the development of gesture recognition algorithms and navigation strategies. It also addresses the various challenges encountered throughout the software's development phase. Furthermore, the report outlines potential enhancements to refine the software's functionality, aiming for optimal performance in real-world applications.

Simulated robot

Using Stage Ros, the robot is able to between waypoints associated by table number order from camera by hands. We are able to implement object avoidance without using move_base library.

Hand Gesture Recognition

Showing specific hand gestures, (zero to five) allows to send order to robot which table number to go to. This mainly relies on two libraries: Mediapipe Machine Learning library developed by Google and OpenCV Open source Computer Vision library (for real time hand detection).

BibTeX

No Bibtex is included since this is not a published paper.