LLM directly edits π0.5 VLA weights, no fine-tuning, lifting robot task success from 30% to 90%
Rewkang · x · 2026-11-18
Abdul Rahim Mirani had Claude Opus 5.5 directly edit the weights of the public π0.5 DROID policy — no training, no fine-tuning. After watching failures in a simulated colour-sorting task (Isaac Sim, Franka arm, Robotiq gripper), the model diagnosed the cause and turned up four attention heads 3×, changing just 0.05% of weights. Successes rose from 9/30 to 27/30 (30%→90%), and the cups-in-basket task — never used to design the edit — improved from 20/30 to 27/30 (67%→90%). All before/after numbers come from 30 unseen object placements per task. The human set the question and reviewed results; Opus 5.5 watched failures, designed and ran every experiment.
More from coding & agent
- GLM-5.3 pricing called absurdly cheap: $10 plan reportedly unlocks $500 in usage — rez0__ · 2026-11-18
- An entire photorealistic video rendered purely with code — no video model, no Blender — JeremyNguyenPhD · 2026-11-18
- Pawel Huryn open-sources 45 agent skills distilled from shipping three live products — PawelHuryn · 2026-11-18
- Code model eval: Rust hits 95-100% accuracy as 13 languages see big gains — shuding · 2026-11-18
- Splash 1.7.0 released: faster decoding for coding agents on Apple Silicon — wojtek15 · 2026-11-18
- Law school AI course wraps: 6 student teams ship working legal AI prototypes — jkubicki · 2026-11-18