Dynamic Reconfiguration of Mission Parameters in Underwater Human-Robot Collaboration.

This paper presents a real-time programming and parameter reconfigurationmethod for autonomous underwater robots in human-robot collaborative tasks.Using a set of intuitive and meaningful hand gestures, we develop asyntactically simple framework that is computationally more efficient than acomplex, grammar-based approach. In the proposed framework, a convolutionalneural network is trained to provide accurate hand gesture recognition;subsequently, a finite-state machine-based deterministic model performsefficient gesture-to-instruction mapping, and further improves robustness ofthe interaction scheme. The key aspect of this framework is that it can beeasily adopted by divers for communicating simple instructions to underwaterrobots without using artificial tags such as fiducial markers, or requiringthem to memorize a potentially complex set of language rules. Extensiveexperiments are performed both on field-trial data and through simulation,which demonstrate the robustness, efficiency, and portability of this frameworkin a number of different scenarios. Finally, a user interaction study ispresented that illustrates the gain in usability of our proposed interactionframework compared to the existing methods for underwater domains.