---
title: "Dyna Robotics releases world-action model trained on 1M hours of video"
url: https://www.parallelquant.com/posts/dyna-robotics-releases-world-action-model-trained-on-1m-hours-of-video-c1265e
source_name: "MarkTechPost"
source_url: https://www.marktechpost.com/2026/08/13/dyna-robotics-introduces-dyna-2-a-world-action-model-pre-trained-on-1-million-hours-of-human-video/
published: 2026-08-13T07:42:02.000Z
topics: ["robotics", "research"]
publisher: "Parallel Quant"
---

# Dyna Robotics releases world-action model trained on 1M hours of video

*2026-08-13 · Source: [MarkTechPost](https://www.marktechpost.com/2026/08/13/dyna-robotics-introduces-dyna-2-a-world-action-model-pre-trained-on-1-million-hours-of-human-video/)*

Dyna Robotics released Dyna-2, a world-action model pretrained on more than one million hours of egocentric human video. Its technical report establishes a scaling law for training on human video up to that scale and shows the law transfers to unseen robot data. The company says video co-training drives generalization across different robot embodiments.

**Why it matters:** Robotics has long been bottlenecked by the scarcity of real robot demonstration data compared to the vast supply of human video online. Dyna-2's finding — that scaling laws learned from human video transfer to robot performance — points to a path for training capable robot policies without proportionally scaling expensive teleoperated data collection, echoing how internet-scale pretraining unlocked progress in language models.

**Topics:** robotics, research

---
Read the original: https://www.marktechpost.com/2026/08/13/dyna-robotics-introduces-dyna-2-a-world-action-model-pre-trained-on-1-million-hours-of-human-video/
Canonical: https://www.parallelquant.com/posts/dyna-robotics-releases-world-action-model-trained-on-1m-hours-of-video-c1265e
Published by Parallel Quant — https://www.parallelquant.com