← Back to home02 / Spatial AI

Spatial AI.

My goal is to solve the challenges of 3D vision and teach AI to understand space and work with it in ways that are useful in practice to both people and robots. Then I want to move toward Physical AGI: building systems that can see the physical world and model it, plan actions within that model, and carry out those actions in reality.

In my PhD dissertation, I developed a classical approach to building 3D models from a series of images, based on structure-from-motion and parametric descriptions of objects. We are now taking this approach to a new level using foundation models for geometry, multimodal LLMs, and agents. I think many 3D vision problems will be solved in the next 2–4 years, with many useful products built on those solutions. I want to contribute to this together with my colleagues, PhD students, and students, and then move on to the next challenge.

The two key domains right now are CAD (mechanical engineering) and Indoor (interior spaces).