Runway releases open-weight Praxis-1 robot control model
Original title: Runway's open-source Praxis-1 turns video AI into a robot controller
Weights and code are Apache 2.0; Runway reports 46.2% on RoboLab-120, but those scores are not independently verified.
Runway released Praxis-1, a World Action Model that turns camera images, text instructions, and robot position data into movement commands. It was introduced to select partners in late September and is now available with weights and source code under Apache 2.0. Standard Bots, Noble Machines, and Ultra have tested it on humanoid, industrial-arm, and dual-arm systems. Runway says the base model still needs extra training for new robots and an 80 GB GPU.
Unconfirmed: Runway ran the benchmarks itself, and the results have not been independently verified.
- Partner access began in late September; weights and code are now public.
- Franka adaptation used about 277 hours of public DROID recordings.
- Dual-arm training used 1,400 to 2,600 hours of material.
- On a real Franka arm, Praxis-1 succeeded 88 of 100 attempts versus 80 for FLUX 3 Action.
- A compressed build is estimated to fit an RTX 5090 with 32 GB of VRAM.
- Vendors
- Runway
- Organizations
- Standard BotsNoble MachinesUltra
- AI models
- Praxis-1
From the original article
4 min read at the-decoder.com
Runway releases its robotics model Praxis-1 with open weights and source code under the Apache 2. 0 license. The model uses knowledge from video training for robot control and leads the World Action Model category on Nvidia's RoboLab-120 benchmark.
By Tomislav Bezmalinović · the-decoder.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.