NVIDIA's ADEPT Accepted at CoRL 2026: Sim-Pretrained Dexterity Deploys Zero-Shot on Real Robots
ankurhandos · x · 2026-11-11
ADEPT, NVIDIA Dex team's pre-training + post-training RL paradigm for dexterous manipulation, is accepted to CoRL 2026 (Poster Session III, Nov 11).
- Motivation: reach, grasp, lift, insert and transport skills get relearned from scratch per task; ADEPT pre-trains one foundational dexterous policy in sim via a generic object-reposing task, paying that cost once;
- Post-trains task specialists while preserving pretrained behavior, then distills into perceptive policies from raw vision + touch, deployed zero-shot on real high-DoF arm-hand robots at human speed;
- Efficiency: each new task takes 3B env steps vs 9B from scratch (which rarely succeeds) — a 3× saving;
- A joint-space geometric fabric controller enforces joint limits and collision safety identically in sim and on hardware.
More from Embodied
- Maestro paper: cheap Gemini 2.5 beats π₀.5 baselines at zero-shot robot control — dineshjayaraman · 2026-11-11
- OpenDDE: open-source all-atom drug discovery engine hits 51% on hard antibody-antigen benchmark — teortaxesTex · 2026-11-11
- Tesla demos Optimus walking inside a world model at CoRL, real policy in control — Scobleizer · 2026-11-11
- NUS Proposes WM-PPO: World-Model-Predicted Successor States Improve PPO Policy Learning — NationalUniversityofSingapore · 2026-11-11
- DexForesight Boosts Dexterous Manipulation Success From 49.4% to 60.5% With Implicit Spatial Foresight — Sharpa-Robotics · 2026-11-11
- NTU Study Finds Time-Indexed Memory Makes VLA Robots Mistake Pauses for Progress — NanyangTechnologicalUniversity · 2026-11-11