Runway has released Praxis-1, an AI model that can control real robots, and it is now available for anyone to download and use for free under an open license. This model is special because it learned a lot about the physical world by watching huge amounts of regular videos (the same kind of training that helps video AI systems). It then turns that knowledge into actions a robot can take.
Praxis-1 is a 27-billion-parameter model. It looks at what a camera sees, knows the robot’s current position, and receives a simple instruction (like “pick up the ball and put it in the box”). It then predicts both what the scene should look like next and the exact movements the robot needs to make. Early partners already tested it on different machines, including a walking humanoid robot, a factory-style arm, and a dual-arm warehouse robot.
On a popular robot test called RoboLab-120, Praxis-1 achieved about 46% success, which is better than other open robot models. The company released the model weights, the code to run it, and tools to fine-tune it on your own robot or simulation. You can adapt it with relatively little extra training data. This is useful for students and hobbyists who want to experiment with robot AI without starting from scratch or paying for a closed system. It shows how video-trained AI can help robots handle everyday tasks like moving objects, dealing with soft materials, or working in messy real-world spaces.
Source: runaway.