On the performance evaluation of a vision-based human-robot interaction framework

This paper describes the performance evaluation of a machine visionbased human-robot interaction framework, particularly those involving human-interface studies. We describe a visual programming language called RoboChat, and a complimentary dialog engine which evaluates the need for confirmation based on utility and risk. Together, RoboChat and the dialog mechanism enable a human operator to send a series of complex instructions to a robot, with the assurance of confirmations in case of high task-cost or command uncertainty, or both. We have performed extensive human-interface studies to evaluate the usability of this framework, both in controlled laboratory conditions and in a variety of outdoors environments. One specific goal for the RoboChat scheme was to aid a scuba diver to operate and program an underwater robot in a variety of deployment scenarios, and the real-world validations were thus performed on-board the Aqua amphibious robot [4], in both underwater and terrestrial environments. The paper describes the details of the visual human-robot interaction framework, with an emphasis on the RoboChat language and the confirmation system, and presents a summary of the set of performance evaluation experiments performed both on- and off-board the Aqua vehicle.