VLA Handover: Failure-Prediction-Based Switching from On-Robot to Server VLA Models
VLA Handover: Failure-Prediction-Based Switching from On-Robot to Server VLA Models
This research introduces VLA handover, a method where robots run small vision-language-action models locally and switch to large server-based models only when failure is predicted. Tested in simulation with SmolVLA and pi_0.5 across three of four LIBERO suites, the approach increased success rates by 11-22 percentage points compared to using small models alone, reduced server requests by 49-77% relative to server-only control (with a minor trade-off in success rate), and showed improved robustness to network latency.
The study demonstrates a heuristic switching framework between small and large vision-language-action models that improves success rates in simulation while reducing server request costs, though results are limited to simulated environments and specific model configurations.
Source: arXiv Robotics — research abstracts · Read original article ↗
Loading article text…
What the source reports
Publisher-reported claims, with original evidence. These results have not been independently verified by RoboSignal.
- Environment
- SimulationOpen source S3
- paper: not_reported
Source excerpts and review record
Automatically extracted; no manual editorial approval recorded.
a small VLA (S-VLA) and switches to a large VLA (L-VLA) on a server only when the S-VLA is likely to fail. We formulate the switching as minimizing the request cost under a required success probability and, since this problem cannot be solved directly, propose a heuristic framework that retrains neither VLA. In simulation with SmolVLA and $\pi_{0.5}$ on three of the four LIBERO suites, VLA handover raised the success
Open source S3
rate by 11-22 percentage points over the S-VLA alone and sent 49-77% fewer requests than server-only control at a 2-4 percentage-point lower success rate, and its completion time grew less with the network latency.
Open source S4
Source：arXiv Robotics — research abstracts · arxiv.org