General-purpose robots are unlikely to be deployed with all of the knowledge they need to complete every possible task. We expect that robots will need to learn after deployment, whether to customize their behavior to an individual user, or to learn to handle situations that were not anticipated by the robot designers. In particular, non-expert users in public spaces could be a rich source of information for robots, but the data obtained from these interaction is often noisy and sparse. Our work in this area both enables robots to learn more effectively from non-expert users in noisy real-world environments and equips robots with the social and interaction skills to help non-expert users provide more useful examples.
This work allows robots to quickly and naturally influence, understand, and learn from people in groups and crowds, while making minimal assumptions about the specifics of their behavior. Additionally, we develop fast new algorithms that allow robots to act as lively, responsive, and appealing social partners, while also accomplishing instrumental tasks relating to their embodiment (e.g., driving around a building to show a visitor where to go, or picking up objects to clear them off a table).
Finally, our work is focused on understanding the needs of "non-normative" users, that is people like children, older adults, or people with disabilities, who are not included in the "convenience populations" with which much robotics research is done. Our goal is to use inclusive design and participartory research approaches to gain a better understanding of the needs of these users, and to develop new computational solutions that address real needs at the intersection of physical assistance and social support.
Elaine Schaertl Short
c/o Department of Computer Science
177 College Ave
Medford, MA, 02138