Other than the PI's prior work, the current state of the art in human-robot interaction does not include algorithms which closely tie engagement behaviors with a sophisticated model of task-oriented collaboration. Thus the most "engaging" current humanoid robots typically lack the ability to collaborate on complex tasks. Conversely, current robots with sophisticated task skills are typically stilted and unnatural to interact with. The PI's ultimate goal in this research is to develop new fundamental computational principles and algorithms which will significantly improve the ability of autonomous robots to collaborate with humans in a broad range of situations in the home, in commercial functions, and in hazardous environments. The specific focus of this project is on engagement, the process by which two (or more) participants in a collaboration initiate, maintain and terminate their perceived connection. Examples of generally applicable engagement behaviors include looking, nodding, pointing and body stance. The PI's working hypothesis, supported by his prior work, is that the proper interpretation and generation of these engagement behaviors by a robot is crucial to the overall effectiveness of its collaboration with humans. The research method will combine the study of engagement in human-human collaboration with the implementation and evaluation of algorithms for human-robot engagement and collaboration using an experimental humanoid robot. In the evaluation, a human will first teach the robot a new procedural skill and then the robot will teach the same skill to a different human. The main scientific output of the project will be a broadly applicable algorithm for initiating, maintaining and terminating engagement in the context of human-robot collaboration. For example, this algorithm will include rules which specify, relative to the state of the collaboration, when and where the robot should look, nod, point and face, and how to interpret the corresponding behaviors by humans. A secondary contribution of the project will be to demonstrate a collaboration-based approach to naturally instructing a robot.
Broader Impacts: Better engagement and collaboration abilities will make it easier for robots to assist the handicapped and the elderly, will make robots more effective in search and rescue operations, and will open up new markets for robots in sales and entertainment. The results of this project will be widely disseminated both through scientific publication and free distribution of the implementation code. The experimental humanoid robot designed by the PI for this work is also commercially available at a reasonable cost. This dissemination will facilitate the results being applied by the designers and builders of future human-robot systems.
