My research focuses on building general-purpose robot learning systems. I draw inspiration from how humans learn many skills efficiently, improve themselves autonomously, and cooperate and interact with one another effectively.
One direction I am particularly interested in is the internal model in the cerebellum, which acts as a “world model” of body and environmental dynamics, enabling predictive motor control.
Inspired by this, in the ASAP and IMPACT projects, we let the robots learn internal models of themselves and object dynamics. This enables agile whole-body skills for humanoid robots and adaptive manipulation under forceful interactions.
I believe future AI systems will resemble a Society of Minds.
Therefore, I am interested in designing protocols and algorithms that allow multi-agent AI systems to communicate and cooperate effectively.
In the CooHOI project, we study how multiple humanoid robots can collaborate on object-carrying tasks, and explore the principles of scalable embodied cooperation.