My research aims to replicate human-like spatial intelligence with minimal human supervision, paving the way for trustworthy and accessible autonomous robots and embodied agents working for and around humans. My current efforts are in three dimensions: (1) agent: building a generalist embodied visual agent powered by foundation models to perceive and act in 3D; (2) world: building a photorealistic world simulator to train and evaluate the foundation agents; (3) data: visual-spatial and multimodal dataset curation.