FIND preprint: robot picks its own weaknesses, lifts 8-task success from 55% to 71.9%
GeorgiaChal · x · 2026-10-10
Researchers introduce FIND, an agentic, reset-free self-improvement framework where a robot selects feasible tasks, focuses on weaker skills, evaluates its own attempts, and improves via real-world RL with no human reward labels.
- Success across 8 tasks rises from 55% to 71.9%
- In a representative 6-hour run, FIND completed 456 episodes with only 30 human scene-recovery interventions
- Authors call it a step toward robots that take charge of their own learning
More from Embodied
- Xpeng shows Physical AI at Paris Motor Show, vision-only FSD heading to Europe — LinusEkenstam · 2026-10-10
- Musk: Digital Optimus beats Diablo halfway through with no APIs, just screen pixels — XFreeze · 2026-10-10
- Tongji team builds spiderweb-inspired 6-axis robotic 3D printer with self-supporting structures — lukas_m_ziegler · 2026-10-10
- Robot hand turns book pages: UVTA visual-tactile-action model hits 70% vs 29% baseline — CyberRobooo · 2026-10-10
- Sony surgical robot's autonomous suturing demo sparks debate over medical AI liability — TansuYegen · 2026-10-10
- VioLA trains humanoid policies on 140M mostly-human frames, hitting 88.6% zero-shot manipulation — Michael_J_Black · 2026-10-10