Announcing the World's First Automated Simulation Asset Generator
Today we're opening early access to Palatial V0.1, the world's first automated simulation asset generator.
Every object a robot learns to grasp, open, or carry has to exist in its simulator first. It needs correct collision geometry, mass, friction, and working joints. Teams still author most of this content by hand, often spending days on one object. The environment becomes the bottleneck long before the algorithm does.
We're automating that work end to end.
A photo becomes a working simulation asset
Give Palatial a text description or an image. The pipeline returns clean geometry, precise collision meshes, physical properties such as mass and friction, and articulation when the object has moving parts. A cabinet arrives with doors that open on the correct hinges. A toaster arrives with a lever that actually travels.
The asset loads directly into Isaac Sim with no setup. Before you receive it, we run it in simulation and record the test. The files ship with a validation video, so you can inspect evidence that the asset works instead of debugging a mesh on arrival.
There's no human in the loop between request and simulation-ready output.
Early access, plus a public stress test
You can sign up at dashboard.palatial.cloud, describe an object or upload an image, and generate it.
We're also opening a research preview on X. Tag @PalatialSim with a photo or a description, and the pipeline will generate the asset and reply with what it made. We want real requests from robotics builders, in public, because they'll find the edge cases faster than a private test set.
This is V0.1
The pipeline supports rigid and articulated objects today. Deformables such as cables and clothing aren't supported yet. A generation takes tens of minutes, not seconds. Some outputs will be wrong, and the validation video will make those failures visible.
We're shipping now with the failure evidence attached instead of polishing privately. No other system takes an image or description and automatically returns a validated, physics-complete, simulation-ready asset. V0.1 makes that workflow available now, with its limits exposed.
Scaling robot learning requires more than a few hand-built environments. It requires simulation assets to work like infrastructure, available on demand for every appliance, tool, container, and fixture a robot may touch. Our near-term goal is thousands of simulation-ready objects at your fingertips. Over time, billions.
It starts with a refrigerator door that actually opens.