Jitendra Malik: robots ignoring 3D structure are wasting a valuable signal

JitendraMalikCV · x · 2026-09-21

Berkeley's Jitendra Malik reflects on 3D/4D reconstruction — from photogrammetry and multi-view geometry (Hartley & Zisserman) to deep learning era systems like VGGT and SAM 3D that incorporate recognition-driven shape priors. He argues this directly matters for robotics: many learning-era robotics papers ignore 3D structure, 'wasting valuable signal' in a field where robots operate in the 3D world. He cites his group's long real2sim2real tradition, welcomes renewed interest sparked by Astra, and asks that human researchers get credit, not just AI models.

Related event: Berkeley's Jitendra Malik: Robots Wasting Valuable 3D Structural Signals(2 posts)→

Original post →

More from Embodied

Embodied channel →