Apple's REFACTOR-VLA: Robots Learn Reusable Libraries of Typed Motor Programs
Apple ML Research · rss · 2026-09-02
Apple ML Research presents REFACTOR-VLA. Current VLA models (OpenVLA, π0, RT-2, RDT-1B) are monolithic—they emit raw motor commands without reusable behavioral abstractions, performing poorly on long-horizon tasks and resisting interpretation. The work learns typed motor program libraries unsupervised, tackling the core problem existing approaches like AtomicVLA and AtomSkill avoid: deciding when two action sequences are behaviorally equivalent.
More from Embodied
- Robotics podcast RoboPapers hits 100 episodes, celebrates sim-to-real community growth — ruilong_li · 2026-09-03
- Unity researcher voxelized his living room into billions of millimeter-scale voxels, stunned Stanford VR audience — Scobleizer · 2026-09-03
- Meta's Muse Spark jumps 1.1 to 1.3 in 55 days, photo-to-3D simulation at $0.60 — alexandr_wang · 2026-09-03
- Counterfactual debugging scales sim2real failure diagnosis to 1M steps in world models — sarahcat21 · 2026-09-03
- Mila, Oxford, Cambridge, Tsinghua and 12 more institutions propose ComBodied Agents, a human-centric AI paradigm — jiqizhixin · 2026-09-03
- Agibot's CReF helps humanoids find safe footholds across rough terrain in real time — Scobleizer · 2026-09-03