NVIDIA ADEPT Pre-trains Dexterity for Zero-Shot Robot Deployment
zhengyiluo · x · 2026-08-21
NVIDIA introduced ADEPT, a framework accelerating dexterity via pre-training and post-training using reinforcement learning.
- Method: It first learns foundational dexterity (reach, grasp, lift) in simulation via a generic object reposing task. It then post-trains task specialists while preserving pretrained behavior, distilling them into perceptive policies.
- Efficiency: Post-training takes roughly 3B environment steps vs. 9B from scratch.
- Deployment: Policies deploy zero-shot on real high-DoF arm-hand robots using raw vision and touch at human speed.
- Safety: A joint-space geometric fabric enforces limits and collision safety across sim and real world.
Related event: NVIDIA Unveils ADEPT: Dexterity Pretraining Speeds Up Robot Learning by 3x(3 posts)→
More from Embodied
- Robots enter the tarmac at Japan's busiest airport for a roughly three-year live test — Aiden_Tech_Ai · 2026-10-16
- Team says no technical blockers remain for v1 self-replicating factory using off-the-shelf parts — ihorbeaver · 2026-10-16
- Skip the fancy vision: a $300 automatic screw feeder beats LLM-controlled picking — ihorbeaver · 2026-10-16
- Precise part insertion via LLM-written data collection code and a linear regression aligner — ihorbeaver · 2026-10-16
- Robot fully controlled by frontier LLMs assembles gadgets autonomously for 1.5 hours error-free — ihorbeaver · 2026-10-16
- After 50+ robotaxi rides, user ranks Tesla Cybercab the best rider experience — EricETesla · 2026-10-16