This award is funded under the American Recovery and Reinvestment Act of 2009 (Public Law 111-5).
It is currently unclear what paradigms will be most appropriate for programming autonomous robots. There are many approaches to robot control, such as teleoperation, handcoding, and planning, speech and gesture instruction, each with strengths and shortcomings. Learning from demonstration (LfD) is a compelling alternative in which robots are programmed implicitly from a user's demonstration rather than explicitly through an intermediate form (e.g., hardcoded program) or task-unrelated secondary skills (e.g., computer programming). Current methods either focus on learning tasks with single fixed objectives directly from demonstration data or multivalued tasks where assumed skill-level controllers provide 'symbol grounding' in the form of subtask-level capabilities.
This project focuses on bridging the gap between LfD for single-objective skills and symbol-grounded tasks, and on developing practical algorithms for improved multivalued robot LfD (e.g., through infinite mixture regression). The project is providing more flexible and easier mechanisms for teaching robots new tasks, thus the project is establishing structured pathways for broad populations of society, from secondary schools to research groups, to engage in autonomous robotics. This project further emphasizes the development of standardized, accessible, and reproducible robot platforms, development of transferable undergraduate autonomous robotics courses, and activities for broadening participation in computing.
