LLM-driven arm beats π0.5 on precision tasks 37 vs 9, but loses on box packing
paigeinsf · x · 2026-11-12
Part 3 of a robot benchmark thread: Fable, a language-model-driven robot arm, clearly beat fine-tuned π0.5 on precision tasks — 37 vs 9 on plugging in a charger, 26 vs 14.5 on inserting tubes, though absolute scores remain low.
The exception was box packing: π0.5 won 51 vs 39 with every run between 45-60, while Fable swung from 5 to 100. The authors' hypothesis: π0.5 learned those motions from 500 demos on this station and reliably repeats them, whereas Fable plans from scratch each run — brilliant when the plan works, falling apart when it doesn't.
More from Embodied
- Gemini Robotics 2.5 demo opens a DJI Mic case from a nudge and a language prompt — bousmalis · 2026-11-12
- LO-World: unified BEV latent model boosts 4D occupancy +10.9% IoU, LiDAR CD -30.1% — Zhimin Liao · 2026-11-12
- CoRL 2026 to host first live Child-Robot Dexterity Challenge on stage — k7agar · 2026-11-12
- Claude Fable 5.2 tops AA-Robotics v0.1 at 75, first model to pass measure box via video-only sizing — ArtificialAnlys · 2026-11-12
- AI agents drive through GTA 6's Vice City with an army in tow — MicahBerkley · 2026-11-12
- Claude designed a humanoid robot overnight: 698 parts, 33 actuators, self-iterated balance — Distinct-Question-16 · 2026-11-12