Berkeley's AsyncVLA runs VLA navigation in real time, 40% higher success under 6s latency
svlevine · x · 2026-11-09
Sergey Levine's team at UC Berkeley presents AsyncVLA, an asynchronous control framework that decouples high-level semantic reasoning (run on a remote workstation) from a lightweight onboard Edge Adapter for reactive control, fixing the latency problem that breaks VLA control loops. End-to-end finetuning and trajectory re-weighting align the two asynchronous streams. On real-world vision-based navigation tasks with delays up to 6 seconds, AsyncVLA achieves a 40% higher success rate than SOTA baselines with fewer collisions. Paper, code, and checkpoints are open-sourced; Noriaki Hirose also got it running in real time with combined edge + remote inference.
More from Embodied
- Team hacks robot arms to 85 points, wins MakerMods Physical AI Hackathon — IsaiahBallah · 2026-11-09
- Stanford and Berkeley unveil Flow Reversal Steering, turning VLM hints into dexterous robot actions with up to 95-point success gains — svlevine · 2026-11-09
- Tesla Optimus engineer builds GROXIE, a countertop robot assistant powered by Grok — Scobleizer · 2026-11-09
- RLT paper wins CoRL oral: sample-efficient real-world RL to fine-tune VLA models — svlevine · 2026-11-09
- Embodied First Exam: 1,090 embodied tasks stump coding agents, best solves just 32.9% — LINJIEFUN · 2026-11-09
- NVIDIA AI Day Seoul Draws 2,000+ Developers to Showcase Physical AI, Robotics and Agentic AI — NVIDIAAI · 2026-11-09