AI GitHub
RoboCasa for simulated household manipulation
Study manipulation tasks in simulated kitchens with varied scenes and objects.
Key Takeaways
- RoboCasa supplies household task environments and demonstrations for policy learning.
- Pick the task and supported simulator configuration before selecting demonstrations.
- Train or evaluate on a small task and reserve several kitchens for testing.
- Scene variation increases difficulty and data needs.
How it works
RoboCasa supplies household task environments and demonstrations for policy learning. Variation across layouts and objects lets an experiment test whether behavior transfers beyond one scene.
What you need
Pick the task and supported simulator configuration before selecting demonstrations. Record the train and evaluation layouts so the split is reproducible.
What to try first
Train or evaluate on a small task and reserve several kitchens for testing. Inspect failure recordings to see whether errors come from navigation, grasping or sequencing.
Evaluate on a new arrangement with the same high-level goal. A recording of the attempt helps reveal whether the policy learned the task or memorized one kitchen layout.
Where to look for failures
Scene variation increases difficulty and data needs. Keep train and evaluation scenes separate and record the simulator and asset versions used.
Related Franklin pages
Start with saved data or a simulator before deploying a policy on hardware. Verify observation layout, action conventions and task success using a known reference episode.
- Record robot demonstrations and train policies using reusable data and control tools. LeRobot for robot demonstrations and policy training
- Run and adapt published robot policy examples and checkpoints. OpenPI for experimenting with vision-language-action policies
Franklin’s robotics coverage gives examples of where simulation, training and hardware meet. Franklin’s robotics coverage
Project documentation
Try it for your workflow
Explore the project’s documentation for setup and usage, or save this page to a collection for later.
Read the project documentation ↗