ActionPiece rethinks action tokenization for VLA models, hits 94.8% on LIBERO
burny_tech · x · 2026-09-20
The paper ActionPiece: Rethinking Action Tokenization for VLA Models redesigns action tokenization for vision-language-action models, preserving physical relationships between actions via Physical Rank Consistency (PRC) instead of naive discretization.
- 94.8% on LIBERO
- 68.8% on the out-of-distribution LIBERO-Plus
The key idea is encoding physical constraints directly into the tokenizer to improve robot policy generalization.
More from Embodied
- Eric Topol: wearable HRV and readiness scores lack evidence of health benefits — EricTopol · 2026-09-21
- Remote-controlled cockroaches now carry cameras and drug injectors for disaster response — Distinct-Question-16 · 2026-09-21
- Camera-only autonomous drone clears obstacle course with Jev judgment model, built in 15 minutes for $0.10 — alex_verem · 2026-09-20
- Humanoids Summit kicks off in Seoul in two days, spotlight on humanoid robotics — jonstephens85 · 2026-09-20
- GPT-6 Astra for robotics: a curated collection of papers, demos and evals — NielsRogge · 2026-09-20
- Guy wires AI into a Big Mouth Billy Bass fish that gives motivational speeches — scaling01 · 2026-09-20