Runway stated it’s testing Praxis-1 on quite a lot of embodiments and environments to determine and shut potential gaps earlier than transferring to normal availability. | Supply: Runway
Runway AI Inc. this week launched Praxis-1, an open-weight world motion mannequin that turns Runway’s video pretraining into management for robots. The corporate stated it constructed Praxis-1 on the identical large-scale video pretraining behind its normal world fashions.
“Most robotic insurance policies are bottlenecked by robotic information, which is scarce and costly to gather,” Kamil Sindi, Runway’s chief expertise officer, advised The Robotic Report. “Praxis-1 learns largely from third-person video, constructed on the identical large-scale pretraining behind Runway’s world fashions, so it already understands how objects behave and the way duties unfold.
“Additionally, efficiency improves as we scale video, so its ceiling is ready by how a lot video it may well be taught from, not what number of robotic demonstrations exist,” he added.
Based in 2018, Runway AI focuses on generative synthetic intelligence analysis and applied sciences. The company affords a spread of AI fashions, together with Aleph 2.0 for in-context video enhancing, Act-Two for movement seize, and Gen-4.5 for text-to-video and image-to-video technology.
For robotics, Runway additionally affords GWM-1, a normal world mannequin. The Brooklyn, N.Y.-based firm has places of work in New York, San Francisco, Seattle, London, Paris, Tel Aviv, and Tokyo.
Runway has been testing Praxis-1 with early companions and plans to launch the mannequin publicly within the coming months.
Editor’s word: Bodily AI is among the many session matter tracks at RoboBusiness 2026, which will likely be on Oct. 20 and 21 in Santa Clara, Calif. Register now to attend.
Register now and assist us have fun 20 years of RoboBusiness!Runway bets on video information for robotic coaching
Whereas there’s a lack of real-world robotics information to coach generalist AI fashions, there may be an abundance of video information. Every single day, individuals movie and add extra of on a regular basis life every day than any robotic lab may seize by way of teleoperated demonstrations, famous Runway AI.
Runway has discovered that simulating robotic insurance policies inside its world mannequin predicts real-world outcomes with 0.95 correlation. This compares favorably with dearer 3D reconstruction–primarily based strategies, the corporate claimed.
As well as, Runway has prolonged its work on pretraining giant video fashions into interactive, real-time video fashions like Solaris and GWM Worlds 2. By educating fashions generate correct physics, how arms transfer, what a process seems to be like partway by way of, the corporate stated it has created dynamic, advanced environments for agent coaching within the digital and bodily world.
Runway stated Praxis-1 brings the identical method to robotics, offering a generalist coverage mannequin for robotics builders and researchers that works throughout any embodiment or setting.
“Praxis-1 has been skilled on quite a lot of manipulation duties, starting from easy pick-and-place actions, corresponding to lifting soda cans, to extra advanced duties involving deformable objects, like packing present luggage,” Sindi stated.
Praxis-1 is already testing with early companions
Runway is rolling Praxis-1 out to key companions forward of a public launch. The corporate can be hoping to supply entry to further companions pre-launch. Runway plans to judge its mannequin throughout quite a lot of robots.
“Early companions, together with Noble Machines, Commonplace Bots, and Extremely, are operating Praxis-1 on their very own {hardware},” Sindi stated. “An enormous takeaway thus far is {that a} single mannequin can adapt throughout very completely different embodiments, from bimanual arms to humanoids, with gentle fine-tuning. Their testing helps us determine and shut gaps earlier than normal availability.”
When Runway does launch Praxis-1 publicly, it plans to ship it with open weights somewhat than as a closed mannequin.
“We imagine U.S. management in bodily AI is vital to regaining our manufacturing lead, and that requires open American fashions,” Sindi stated. “Open weights give {hardware} builders flexibility and management they don’t have as we speak.”
Wanting forward, the corporate hopes to pre-launch with extra choose companions to proceed enhancing its mannequin forward of its full launch.
“We’ll proceed testing and evaluating Praxis-1 with companions, carry on extra early-access companions, and assess efficacy and security throughout completely different embodiments and environments,” Sindi stated.
Runway says a coverage that already understands bodily plausibility and object conduct from video pretraining has an infinite head begin on one constructed from motion information alone. | Supply: Runway
The submit Runway introduces Praxis-1 world motion mannequin for robotics appeared first on The Robotic Report.
