Columbia robotics researcher Yunzhu Li argues manipulation, not locomotion, remains the core bottleneck for real-world robot deployment

Yunzhu Li, an assistant professor at Columbia University and co-founder of SceniX, specializes in robotics simulation and robot learning. SceniX is developing tools for robotics companies to generate training data, build simulation environments, and evaluate robot performance to accelerate deployment from lab to real-world settings.

Li distinguishes locomotion from manipulation: walking across a room requires a robot to model and control its own body, while manipulation demands understanding of objects, materials, geometry, contact, and how the environment changes through interaction. Handling deformable objects, organizing clutter, and reliably manipulating unfamiliar items in new environments remains much harder because small errors can quickly change task outcomes.

Simulation is especially important for physical AI because collecting real-world robot data is slow, expensive, and difficult to scale safely. In simulation, robots can generate diverse training data, test edge cases, and learn about environments they need to interact with before real-world deployment. Li argues that while robots themselves are relatively easy to model, the real world—with its diverse objects, materials, contact dynamics, and constant changes—is not.