Runway Unveils Praxis-1 Robot Controller

Praxis-1 advances a compelling thesis: general video could make robot training far less dependent on scarce, expensive teleoperation data. The evidence is promising but early, so the real test is reliable transfer across bodies, environments, and safety-critical edge cases.

• –Runway reports nearly identical placement errors for web-video and teleoperation-video pretraining: 16.1 cm versus 16.0 cm across 93 evaluation pairs.
• –Early partners span bimanual systems, a six-degree-of-freedom arm, and a mobile robot base.
• –The reported 0.95 simulation-to-reality correlation measures ranking alignment, not task success or safety.
• –Runway highlights clutter, transparent objects, repeated items, and deformable cloth as key challenges, but provides no category-specific success rates.
• –Open weights could give robotics developers more control than closed policy models, but they are not publicly available yet. [Independent analysis](https://www.superpowerdaily.com/posts/runway-brings-video-pretraining-to-robot-control-with-praxis-1)