# FlowVLA preprint reports robot manipulation gains from 3D motion training

_Published Tuesday, October 13, 2026 at 10:41 PM EDT · Robotics, Science, AI · Latest · Tier 2 — Notable_

Researchers report that FlowVLA improved robot manipulation, zero-shot robustness and data efficiency in simulation and real-world experiments without adding computation at inference. The preprint introduces a training framework that teaches vision-language-action models how task-relevant points move through three-dimensional space.

The method uses reconstructed motion to supervise both point movements and their alignment with robot control trajectories. It adds lightweight auxiliary components during training and removes them for inference, preserving the existing model architecture and requiring no additional 3D sensor inputs. The reported gains come from experimental evaluations.

## Sources

- [arXiv Query: search_query=cat:cs.RO&id_list=&start=0&max_results=30](https://arxiv.org/abs/2610.16160v1)

---
Canonical: https://techandbusiness.org/newswire/l7w92juXVOMlcHIRBJh6Ix
Published: 2026-10-14T02:41:44.642Z
Story chronology: 2026-10-12T04:34:14.000Z
Retrieved: 2026-10-14T05:01:41.389Z
Publisher: Tech & Business (techandbusiness.org)