ai/list101

Runway Praxis-1

runway.com

An open-weight robot control model that combines video prediction and action generation for adaptable robotics policies.

roboticsDeveloper Tools
Visit website
Screenshot of Runway Praxis-1 official website
Pricing
Free
Available on
Self-hosted, Python, NVIDIA CUDA
Category
robotics

What is Runway Praxis-1?

Runway Praxis-1 is a world action model for robotics developers and researchers. It applies video pretraining to robot control, taking camera observations, robot state and a task instruction to predict actions.

The release includes model weights, inference code and fine-tuning code under Apache 2.0, including commercial use. Developers can run and adapt the policy on their own hardware, with full fine-tuning and LoRA available for customization.

Key features

Joint video and action prediction

A 27B-parameter mixture-of-experts diffusion transformer predicts video and robot actions together.

Robot-specific checkpoints

The repository documents DROID single-arm, YAM bimanual and SO-100/SO-101 variants, with checkpoint-specific observation requirements.

Fine-tuning options

Full fine-tuning supports broader adaptation, while LoRA provides a documented route for task adaptation.

Simulation and policy serving

The release supplies a policy server and MuJoCo simulation for developing and evaluating robot policies.

Robotics ecosystem integrations

Distribution includes GitHub and Hugging Face, with native LeRobot support and vLLM-Omni integration.

Who is it for?

Researchers evaluating manipulation policies

Compare task behavior in simulation before planning a physical robot evaluation.

Robotics teams adapting an existing rig

Select an appropriate checkpoint and investigate adaptation to the rig's observations and controls.

Developers specializing a task

Explore LoRA adaptation for a narrowly defined manipulation workflow.

How to use Runway Praxis-1

  1. Review the repository's hardware requirements and checkpoint configurations.
  2. Prepare Python 3.12, the documented CUDA environment and a compatible GPU.
  3. Clone and install the repository, then authenticate with Hugging Face and download the required weights.
  4. Run the single-arm simulation quick start before exploring rig adaptation and evaluation.

Runway Praxis-1 pricing

Free

Weights, inference code and fine-tuning code are available under Apache 2.0, including commercial use. Self-hosting requires separately supplied compute and robot hardware; the open release does not include those resources.

Check current pricing ↗

What to know before you use it

  • The default quick start requires a CUDA GPU with at least 80 GB memory and a CUDA 13.0 driver. This requirement applies to that documented workflow.
  • Camera arrangements and robot state inputs depend on the checkpoint; deployment requires matching or adapting the target setup.
  • Runway reports failures involving look-alike objects and vague instructions. Naming distinguishing attributes helps identify the intended target.
  • Chaining multiple steps from one prompt remains difficult. The research report describes better results when each step receives its own instruction.

Frequently asked questions

Can Praxis-1 be used commercially?

Yes. Runway releases the weights and inference and fine-tuning code under Apache 2.0, including commercial use.

Can evaluation begin without a physical robot?

Yes. The repository's quick start runs a single-arm checkpoint in simulation, subject to its compute and software requirements.

What does a policy call return?

The documented single-arm policy returns robot commands. Predicted video for the corresponding time span is available when video decoding is enabled; serving configurations that skip decoding still return robot commands.