Berkeley's Angjoo Kanazawa teaches machines to understand the physical world for the robotics era
jfiance · x · 2026-09-17
Angjoo Kanazawa, assistant professor at Berkeley EECS and head of the Kanazawa AI Research Lab within BAIR, works at the intersection of computer vision, graphics, and machine learning: teaching machines to understand people, animals, and 3D environments from ordinary images and video.
Questions humans solve automatically — where a person is positioned, how their body moves, handling occlusion, tracking dynamic scenes — remain hard technical problems. As AI moves into robotics, autonomous systems, and mixed reality, these perception skills become critical: robots get no structured prompt and must infer depth, motion, and uncertainty while continuously updating their understanding. The author notes the current robotics/physical-AI startup wave rests on years of foundational research at Berkeley, with Kanazawa among those pushing it forward.
More from Companies & People
- Stanford's Clinical Mind AI uses simulated patients to teach clinical reasoning at scale — StanfordHAI · 2026-09-18
- Stanford HAI's Fall Seminar Series Returns With 7 Sessions on AI in Work, Health, and Finance — StanfordHAI · 2026-09-18
- Mila and Germany's DFKI partner on trustworthy AI, agents and AI safety — Mila_Quebec · 2026-09-17
- Microsoft claims 20% higher close rates for frequent Copilot sellers — but no baseline — HaktanSuren · 2026-09-17
- Palantir CEO Karp: discounted tokens exist to "steal your alpha" and IP — eliano · 2026-09-17
- Researcher Manuel Faysse joins Google DeepMind to work on Gemini models — antoine_chaffin · 2026-09-17