AI BriefingRunway Unveils Praxis-1, a Robot Policy Model Pretrained on Web Videos
·2026.10.30 08:00
Key point
Runway announced Praxis-1, a robot policy model pretrained on web videos, and previewed an open-weight release within months.
Details
- Runway announced Praxis-1, a robot policy model pretrained on web videos, emphasizing that it is a general-purpose policy model operating across diverse robot morphologies and environments
- When the pretraining video duration was increased from 10 hours to 1,000 hours, the final placement error of the web-video-based policy decreased from 18.2cm to 16.1cm, suggesting that the volume of video learning is a bottleneck for performance
- Model weights are scheduled to be released as open weights within months, but currently, the weights, license, model size, and paper are all undisclosed, and there is no repository on Hugging Face
- Testing is underway on early partner hardware such as Noble Machines, Standard Bots, and Ultra, and pre-release usage requires a selective application to the Runway robotics team
- The purpose of the open-weight release is presented as strengthening US physical AI competitiveness and recovering leadership in manufacturing
This summary was generated automatically by AI. Check the original for the author's claims and context. Copyright belongs to the original author.
Our guide explains how the AI works. Report summary errors, attribution issues, or removal requests via Contact.
Related
Runway introduces Praxis-1, an open-weight world action model for robotics
AWS Releases Guide for Building Physical AI Training Pipelines Using SageMaker and IoT Greengrass