Oopsie Data: multi-lab effort to build first large-scale robot failure dataset
micoolcho · x · 2026-10-22
Robo Papers is teasing an episode on Oopsie Data, a community-driven multi-lab project building the first large-scale dataset of real robot manipulation failures. Key thesis: today's robotics datasets contain only successes, so policies never learn what a bad grasp looks like; real failures and suboptimal behavior are the missing ingredient for RL, reward modeling, failure prediction, and world modeling, and synthetic noise injection is no substitute. Labs contributing rollout data (failures and successes alike) get early dataset access and can qualify for co-authorship on the public release.
More from Embodied
- JHU/MIT OpenTAC tactile skin paper already integrated into tendon-driven robot hands — anand_bhattad · 2026-10-22
- OpenTac-Xampling adaptive sampling hits ~1kHz on 1024-taxel sensor, 15x faster than raster — anand_bhattad · 2026-10-22
- Hobbyist builds a working computer vision system with Clairvoyance and ChatGPT — draginol · 2026-10-22
- LAGEN: one frame of latency can cut real-time agent performance by 95% — wzenus · 2026-10-22
- Humanoid robot fights spark debate: entertainment is the only real use case today — KyleMorgenstein · 2026-10-22
- Unitree H2X humanoid robot sports dual battery slots in hands-on shots — micoolcho · 2026-10-22