We study techniques that use world models to predict the state of unobserved spaces and plan, in real time, exploration paths that efficiently find target objects specified in natural language.
Embodied AI agents
We study learning methods that enable embodied agents in virtual or real environments to perceive their surroundings, set their own goals, act autonomously, and adapt to new situations through experience.
Computer Vision and Augmented Reality
Wide-area camera tracking and 3D reconstruction
Using video captured with mobile devices, we study techniques that track the camera’s position and orientation in real time and reconstruct the 3D structure of the surroundings at large-scale industrial sites.
3DGS+SLAM
We study techniques that speed up SLAM mapping by efficiently placing initial Gaussians using image edges and local structure, reducing optimization cost while maintaining rendering quality.
Computer Graphics
Ray/Path tracing
We study methods that generate photorealistic 3D images by simulating physical light interactions such as reflection and transmission, and combine them with denoising to improve the quality and efficiency of real-time rendering.
Physics-based simulation
We study simulation techniques that accurately and efficiently reproduce a wide range of physical phenomena, including object deformation and motion, fluid flow, and collisions between objects.
