# VLA-Fabric boosts multi‑arm robot success by up to 40%

> The new VLA-Fabric method lets each robot arm run its own vision‑language‑action model while sharing learned interactions, raising task success rates 16–40 points.

Oossa · 2026-10-19 · https://oossa.com/en/vla-fabric-boosts-multi-arm-robot-success-by-up-to-40

Researchers led by Shuaijun Liu released VLA-Fabric, a system that gives each arm its own vision‑language‑action (VLA) model and adds a learned interaction layer between arms. Tests on three‑ and four‑arm setups showed the coordinated version succeeded 16–40 percentage points more often than independent policies. The approach works both in simulation and on real robots, and it is not tied to a specific model architecture.

## The facts

- Full interaction improves success by 16–40 percentage points on four physical tasks.
- Paper submitted on 16 Oct 2026 (published 19 Oct 2026).

## Why it matters

Improved coordination means factories could use fewer, more flexible robot arms for complex assembly tasks.

## Sources & references

1. [VLA-Fabric: Learning Interaction for Composable Multi-Arm VLA Agents](https://arxiv.org/abs/2610.23018) – arXiv cs.RO, 2026-10-19

Last updated: 2026-10-19