Robotics foundation models get a serving layer built for factories
A new paper describes a serving system for robotics foundation models in robot factories — the unglamorous infrastructure question of how you actually run these policies across many machines, with versioning, latency guarantees and failure handling. It is the sign of a field crossing from research into operations.
Serving systems appear when a technology stops being demonstrated and starts being depended on. Language models got theirs years ago and it turned out to be where most of the practical difficulty lived. Robotics reaching the same milestone says more about the field's maturity than any benchmark.
The requirements are also harder here. A language model serving system can queue a request. A policy running a physical arm cannot, because the world does not pause while inference completes, and a late action is frequently worse than no action at all.
Related work on thermal-perceptive models for safety-critical manipulation and on gaze-conditioned policies points the same direction: the research questions are increasingly about deployment conditions rather than capability ceilings.
arXiv — ROSA: a robotics foundation model serving system for robot factories → · arXiv — Safe-Night VLA: thermal-perceptive vision-language-action models for safety-critical manipulation → · arXiv — Gaze2Act: gaze-conditioned vision-language-action policies for interactive robot manipulation →