# PDRVLA researchers report robot control without generating future videos

_Published Tuesday, October 27, 2026 at 10:30 PM EDT · Robotics, AI, Science · Latest · Tier 2 — Notable_

Researchers introduced PDRVLA, a model that generates robot actions from images and language instructions while incorporating physical dynamics without generating future videos during operation. Their preprint reports success rates of 99.3% on LIBERO, 89.9% on LIBERO-Plus, 93.6% on RoboTwin 2.0, and 81.5% on SimplerEnv.

A lightweight module extracts representations of physical dynamics to guide action generation. It learns by predicting future image, depth, and motion features, with additional supervision from a pretrained video world model. Training uses entirely publicly available data, expanded through synthesis and annotations. The reported task success results cover four simulation benchmarks.

## Sources

- [arXiv Query: search_query=cat:cs.RO&id_list=&start=0&max_results=30](https://arxiv.org/abs/2610.37381v1)

---
Canonical: https://techandbusiness.org/newswire/o4iV-CBGY0keOkkRQItDYV
Published: 2026-10-28T02:30:50.953Z
Story chronology: 2026-10-27T15:09:10.000Z
Retrieved: 2026-10-28T04:44:50.074Z
Publisher: Tech & Business (techandbusiness.org)