Introduction: The era of intelligent robotics is written in video
The rapid evolution of robotic systems today hinges on a critical data source: visual experience. Traditional models struggle to generalise beyond the lab because they lack diverse, expensive-to‑collect data. Runway AI, a pioneer in video pre‑training, offers a disruptive solution with its new Praxis‑1 model.
Runway AI and the birth of Praxis‑1
Built on Runway’s expertise in “world” models, Praxis‑1 is an open‑weight action model that leverages billions of video frames to learn real‑world dynamics. Using the same architecture as its vision models, it naturally incorporates object and interaction understanding.
Turning video into robotic control: the heart of Praxis‑1
Unlike supervised approaches that require robot‑specific annotations, Praxis‑1 exploits third‑person video. It infers intentions and action sequences directly from human or animal videos, then translates them into robot‑adapted commands.
A borderless learning experience
The algorithm analyses the kinematics of every object: trajectory, speed, spatial constraints. This capability lets a single model handle varied tasks, from cleaning a table to complex workshop assembly.
Major benefits: cost and performance
Drastically reduced need for robot‑specific data translates into significant cost savings. Moreover, more videos lead to better results, as the model refines its understanding at scale.
- Lower cost: fewer manual annotations
- Increased flexibility: adapts to different robots and environments
- Scalable performance: continuous improvement with more video data
Impact on developers
Engineers can now deploy robotic agents without a lengthy training process. Aligning the model to the target platform requires only minor adjustments.
Testing and validation in real environments
Runway tested Praxis‑1 across several “embodiments”: robotic arms, drones, and autonomous vehicles. Results show a 40% reduction in training time compared to conventional methods.
“Praxis‑1 paves the way for smarter and more accessible robots,” says Kamil Sindi, Runway CTO.
Future outlook: towards democratised robotics
With its open‑weight architecture, open‑source communities can contribute to continuous model improvement. The goal is to create an ecosystem where every new video enriches the collective knowledge base.
A call for developers and researchers
We invite AI robotics enthusiasts to test Praxis‑1, share feedback, and collaborate on innovative use cases. Together, let’s shape a future where robots learn from the world just like we do.
Conclusion: A decisive step toward autonomous robotics
Runway AI, through Praxis‑1, transforms how robots acquire skills. By turning video into concrete control, it opens new possibilities for intelligent automation and industrial innovation.
