Understanding Human Motion and Gestures for Underwater Human-Robot Collaboration

TL;DR

This paper introduces robust visual detection and tracking algorithms for underwater robots to follow divers and a simplified hand gesture communication system, validated through extensive field experiments and user studies.

Contribution

It presents novel diver-following algorithms and a user-friendly gesture-based communication framework for underwater human-robot collaboration.

Findings

Effective diver-following algorithms demonstrated in real-world tests

Gesture recognition framework is accurate and easy to adopt underwater

User study shows improved usability over existing methods

Abstract

In this paper, we present a number of robust methodologies for an underwater robot to visually detect, follow, and interact with a diver for collaborative task execution. We design and develop two autonomous diver-following algorithms, the first of which utilizes both spatial- and frequency-domain features pertaining to human swimming patterns in order to visually track a diver. The second algorithm uses a convolutional neural network-based model for robust tracking-by-detection. In addition, we propose a hand gesture-based human-robot communication framework that is syntactically simpler and computationally more efficient than the existing grammar-based frameworks. In the proposed interaction framework, deep visual detectors are used to provide accurate hand gesture recognition; subsequently, a finite-state machine performs robust and efficient gesture-to-instruction mapping. The distinguishing feature of this framework is that it can be easily adopted by divers for communicating with underwater robots without using artificial markers or requiring memorization of complex language rules. Furthermore, we validate the performance and effectiveness of the proposed methodologies through extensive field experiments in closed- and open-water environments. Finally, we perform a user interaction study to demonstrate the usability benefits of our proposed interaction framework compared to existing methods.