NVIDIA's TAMPAgent uses factor graphs to let robots zero-shot mobile manipulation in unseen homes
danfei_xu · x · 2026-11-05
TAMPAgent (NVIDIA Research / Georgia Tech / Utah, CoRL 2026) tackles long-horizon mobile manipulation under partial observability — targets hidden in fridges or blocked by objects — zero-shot, without simulation rehearsal. Factor graphs give frontier agentic models a planning abstraction: the agent acts as a belief-space policy whose parameterized actions carry constraints that tools must certify before anything moves. Object-location beliefs evolve from uniform priors to language hypotheses to perception-certified masks and RGB-D estimates. Compared against CaP and GaP on nine kitchen tasks (success rate and token use), and validated on a real RB-Y1 mobile manipulator and a Unitree G1 humanoid, replacing costly trial-and-error with structured reasoning.
Related event: NVIDIA's TAMPAgent Enables Zero-Shot Mobile Manipulation via Factor Graphs(2 posts)→
More from Embodied
- AI smartwatch detects hypertension with 66% sensitivity, 90% specificity, Nature study finds — EricTopol · 2026-11-06
- Apple shows off a polished humanoid robot research project page — carlosdponx · 2026-11-06
- NVIDIA unveils VFlash: parallel decoding of visual plans for physical AI — ZhidingYu · 2026-11-06
- NVIDIA runs four hands-on physical AI training labs at SIGGRAPH Asia 2026 — tomaszbednarz · 2026-11-06
- Open-source clayview rebuilds Tesla-style clay 3D street views from dashcam footage — theteknosaur · 2026-11-06
- Aina launches $399 smart ring that surfaces proactive AI cues and agent approvals — kimmonismus · 2026-11-06