UniWALM: one diffusion video backbone for planning, memory and manipulation
drsrinathsridha · x · 2026-11-10
UniWALM is a unified video backbone that plans, remembers, decides, and acts in a single diffusion pass, trained on action + VQA data. Reported results: bimanual memory jumps 11%→61%, RMBench M(n) 34.5% vs Mem-0's 28.5%, Pick & Place +13.4 pts, with predicted subtasks voiced via TTS — a notable take on video models as embodied-AI foundations.
More from Embodied
- Musk: Cybertruck Orders Exceed Production, Line Shuts in ~8 Months as Focus Shifts to Cybercab — minchoi · 2026-11-10
- CoRL 2026 show floor: Dyna zero-shot mobile manipulation, NVIDIA GR00T 2, and more — chris_j_paxton · 2026-11-10
- Beff Jezos wants hands-free AR glasses for navigation after traveling in Japan — beffjezos · 2026-11-10
- NVIDIA DLSS 5 Wins CES 2027 Best of Innovation Award for 3D-Guided Neural Rendering — ctnzr · 2026-11-10
- China's brain-reading device hits 94% mind-typing accuracy in day-long use — yogthos · 2026-11-10
- Brown PhD candidate Rahul Sajnani seeks robotics world-model research roles, advisor endorses — drsrinathsridha · 2026-11-10