Chuchu Chen

My research aims to give embodied systems a physically grounded, predictive understanding of themselves and the world. I approach this goal through probabilistic state estimation and robot dynamics, studying how robots infer the physical states they cannot directly observe and predict how those states evolve through motion and interaction.

I am particularly interested in egocentric perception and contact-rich interaction. I study the robot and its environment as a coupled dynamical system: the robot’s actions change the world, while its onboard observations reveal only part of that evolving interaction. My goal is to develop representations and inference methods that connect sensing, dynamics, and uncertainty—enabling robots not only to estimate what is happening, but to anticipate the consequences of their actions and recognize the limits of their understanding.

This agenda builds on my work in consistent and efficient visual–inertial estimation and SLAM, while extending toward the broader foundations of physical AI. A central setting is brownfield robotics, particularly existing warehouses, where autonomy must coexist with established infrastructure, human activity, and ongoing operations. My long-term vision is robots that learn to understand and work within the physical world, rather than requiring the world to be engineered around them.