CAREER: AI-based Interactive and Adaptive Robotic Teaching Systems for Personalized, Curriculum-Based Human Motor Skill Training

Robots are increasingly capable of helping people complete physical tasks, but many current systems are designed to assist with task execution rather than teach people to develop independent motor skills. Existing robot-assisted training systems can guide movement, provide physical support, or improve short-term performance, but these forms of assistance do not fully address how human learners acquire, retain, and generalize motor skills over time. This Faculty Early Career Development Program (CAREER) award aims to address this gap by advancing robotic systems from task-oriented assistants toward instructional robots that systematically foster human motor skill learning through personalized, adaptive, and interactive teaching strategies. The resulting methods could make physical skill training more accessible and individualized in settings where expert instructional support is limited, costly, or otherwise difficult to obtain, with long-term benefits for rehabilitation, sports coaching, workforce training, and lifelong motor learning. The project will also integrate research with education through new course materials, student mentoring, undergraduate research opportunities, and outreach activities that increase participation in robotics and human-centered artificial intelligence, especially in rural communities. Over the five-year project, the research will develop an adaptable robotic teaching framework for personalized and curriculum-based motor skill training. The work will investigate three connected technical questions. First, it will create generative task modeling methods, including diffusion-based models and structured skill representations, that learn from expert demonstrations and synthesize curriculum-aligned training tasks tailored to each learner’s progress across stages of learning. Second, it will develop probabilistic inference methods for estimating a learner’s evolving skill state from interaction data, including movement trajectories, task performance, control inputs, and force-related measures. These estimates will be used within an adaptive shared autonomy framework and a human-in-the-loop optimization approach to personalize robotic assistance in ways that improve learning efficiency. Third, the project will develop interactive policy adaptation methods that allow human experts to guide and refine robot teaching behavior through intuitive inputs, such as task goals, corrections, or few-shot demonstrations. The integrated system will be evaluated through studies with human participants in controlled motor learning testbeds, including movement control, force and impedance interaction, and remote robot operation tasks. These studies will assess skill improvement, retention, transfer to related tasks, personalization effectiveness, and participant experience. The project will contribute to new computational foundations for robots that promote human learning rather than only assist with task execution. This award reflects NSF's statutory mission and has been deemed worthy of support through evaluation using the Foundation's intellectual merit and broader impacts review criteria.