PaceVLA: per-chunk adaptive planning makes VLA robots up to 1.8x faster at same success
chris_j_paxton · x · 2026-10-28
PaceVLA lets a VLA pick, per action chunk, how coarsely and how far ahead to act, running up to 1.8x faster than teleoperation-paced baselines at equal success rate.
More from Embodied
- Tesla Cybercab Texas registrations hit 463+, on pace for 1,500-2,000 by year end — JOBhakdi · 2026-10-28
- 3D AI startup SpAItial opens 15 roles spanning world models, robotics and 3D reconstruction — risingodegua · 2026-10-28
- NVIDIA ships 2,575 sim-ready 3D assets on Hugging Face as Polyfork opens 2,600+ free models — Vjeux · 2026-10-28
- 5-Year-Old Fetch Robot Gets a LEAP Hand for Dexterous Manipulation — YuXiang_IRVL · 2026-10-28
- Shanghai action plan targets 8,000 humanoid robots in factories by 2028 — pstAsiatech · 2026-10-28
- Purdue's IDEAS Lab demos GPS-free drone that tracks runners through dense forests on-chip — DavidSKrueger · 2026-10-28