NVIDIA's Recova turns robot failures into data via agent-guided recovery
ZhidingYu · x · 2026-10-14
NVIDIA, with UCSD and UT Austin, released Recova to scale robot learning without scaling human labor, by turning failures into training data:
- Real2Sim2Real: agents explore failures in digital twins, where trial and error is safer and cheaper, and author recovery skills.
- Autonomous recovery: robots learn to recover from unexpected failures instead of waiting for human resets.
- Human-on-demand: multiple robots run in parallel; humans step in only when autonomous recovery falls short.
The loop: rebuild the scene, train task and recovery policies on digital-twin rollouts, then run real robots and collect DAgger-style data — extracting more learning per hour of human effort. Demos include mahjong.
Related event: Recova teaches robots to self-recover, cutting human interventions to zero(2 posts)→
More from Embodied
- RigidFormer accepted at NeurIPS: mesh-free Transformer beats GNN simulators for rigid dynamics — camenduru · 2026-10-14
- Mind Robotics, Rivian's robotics spinoff with $1B raised, hires ex-Ford EV chief Doug Field — Sethwinterroth · 2026-10-14
- AlphaNIST robot training: Newton physics plus sparse reward did the heavy lifting — milesmacklin · 2026-10-14
- Dyna-2 Deep Dive: World-Action Models Trained on 1M Hours of Human Video, 99.4% 24-Hour Autonomy — Latent Space · 2026-10-14
- Nature Will unveils Natura-1: 50B-parameter MoE robot foundation model trained on 1M hours of egocentric human video — rohanpaul_ai · 2026-10-14
- 15 open-source ESP32 AI gadget projects plus a step-by-step Meta Muse SDK flashing tutorial — alexcovo_eth · 2026-10-14