Fei-Fei Li's Stanford team unveils I3L, boosting long-horizon robot success rates by 68%
drfeifei · x · 2026-10-31
Stanford's Fei-Fei Li group, with Northwestern and Lambda, introduced I3L (Iterative Interactive Imitation Learning), targeting the "last-mile" reliability gap in long-horizon whole-body mobile manipulation.
Key points:
- The open-source, low-cost JoyLo+ interface lets operators correct a robot's arms, torso, and mobile base mid-deployment, using policy-tracking leader arms and touch-triggered takeover.
- Corrections are fed into iterative retraining rounds, each targeting failures of the updated policy.
- Evaluated across 3 robot embodiments and 8 simulated and real tasks, including 4 long-horizon mobile manipulation tasks.
- Under a four-round HG-DAgger protocol, I3L improves autonomous success over base policies by an average of 68%.
Paper, code, models, and datasets are open-sourced.
Related event: Fei-Fei Li's team releases I3L, boosting long-horizon robot success by 68%(2 posts)→
More from Embodied
- Real2Sim2Real is hard: even plugging in a network cable demands high sim fidelity — jonstephens85 · 2026-10-31
- Genspark unveils VoicePad, a $99 five-key AI voice gadget now open for pre-orders — Scobleizer · 2026-10-31
- World's first humanoid robot factory for clothing launches, mass-producing Annie robots — CyberRobooo · 2026-10-31
- Robotics benchmarks feel impossible; Drift builds evals for real deployment sites — jonstephens85 · 2026-10-31
- Tesla's We, Robot promises materialize: Cybercab in production, Robotaxi in six cities — XFreeze · 2026-10-31
- SO-101, AI's 'most famous and infamous' robot, gets official support — jonstephens85 · 2026-10-31