Runway Robotics

What it does

Runway Robotics applies the company's GWM-1 General World Model to robot learning. It takes camera observations, action sequences and existing training trajectories, and produces predicted robot actions, simulated policy rollouts and augmented training data. The platform is offered as separate components: a policy model, video model licensing, offline policy evaluation and data augmentation.

Teams can use it at different points in a robot learning pipeline. The policy model predicts actions from live camera observations and is fine-tuned to a customer's hardware, environment and tasks. Offline evaluation lets you submit action sequences with camera observations and simulate rollouts in GWM-1 to surface failure modes before deploying to hardware. Data augmentation recasts existing trajectories into new environments, lighting conditions and object configurations. One world model backs every component, and GWM-1 can also be licensed as the diffusion backbone for training your own policy models on your own infrastructure.

It suits robotics and robot-learning teams that want to extend an existing pipeline rather than build everything in-house. Access begins with a request through the Get Access button, and Runway works with you to fit the components to your hardware. Policy inference and evaluation need no specialized compute to get started. Request a quote for pricing, and confirm supported robot types and technical requirements directly with Runway.

What it can do

• Predict robot actions from live camera observations with a policy model fine-tuned to your hardware and tasks
• License GWM-1 as the diffusion backbone for training custom policy models on your own infrastructure
• Simulate policy rollouts from submitted action sequences and camera observations to surface failure modes
• Transform existing trajectories into new environments, lighting conditions and object configurations
• Run policy inference and evaluation without specialized compute infrastructure to get started

Example workflows

Pre-deployment policy check

Inputs: candidate action sequences and camera observations. Steps: submit them for offline policy evaluation, simulate rollouts in GWM-1, and review the failure modes. Output: a list of issues to fix before running on hardware.

Expanding a training set

Inputs: existing robot trajectories. Steps: apply data augmentation to place them in new environments, lighting conditions and object configurations. Output: a broader training distribution without new data collection.

Before you choose

• Request a quote for Robotics pricing; Runway's public creative subscription plans are a separate product.
• The policy model is custom fine-tuned per hardware and environment, so setup involves working with Runway rather than self-serve use.
• Licensing GWM-1 for your own infrastructure means you manage deployment and fine-tuning yourself.

Listed under

• Coding
• Sandboxes and infrastructure

Get started

Request access through the product page; Runway works with you to fit it to your hardware.

1. Select Get Access on the Runway Robotics page to request access.
2. Discuss your hardware, environment and tasks with Runway so the components can be fitted to them.
3. Choose which component you need first: policy model, video model licensing, offline evaluation or data augmentation.

Runway Robotics has its own access process. Creative video subscription prices do not establish the price of robotics models or deployment.

We couldn’t find published prices for Runway Robotics. See their pricing page ↗

Sources for this overview

Product pages, documentation, and pricing checked 28 Oct 2026. Each source supports part of the overview; availability and prices can change.

• Runway Robotics | Build smarter robots with general world models ↗runway.com
• AI Image and Video Pricing from $12/month | Runway AI ↗runway.com

Company website

runway.com: product, pricing and policy pages

Web search

Confirms the official site

Social accounts

Accounts the company links, and its posts on X

Human review

Checked by a person

knowledge_base

Last checked 28 Oct 2026. Every field keeps its source and date; unknown means we couldn’t confirm it.