Palatial V1.0: The First Newton Asset Generator, Deformables Included
Today we're releasing Palatial V1.0, the first automated pipeline to generate sim-ready assets at scale.
Since early access opened in February, we shipped articulated generation at scale in March with V0.2, MuJoCo export in April for the first Y Combinator robotics hackathon, and the Palatial Library on April 21. The pipeline generated more than 10,000 assets across homes, warehouses, factories, and construction sites. That run exposed the failure modes we needed to fix before calling the platform 1.0.
The agent configures the work
V1.0 puts an agent in charge of the pipeline. It reads your context, determines what the asset is and how it should behave, then configures the generation settings automatically.
Inputs are multimodal. You can provide text, images, CAD files, or datasheets. The release also includes a rebuilt interface with organization support, so teams can share queues, assets, and results instead of working from separate accounts.
Newton, including soft bodies
Newton is the open-source physics engine initiated by Disney Research, Google DeepMind, and NVIDIA. It is GPU-native, OpenUSD-first, and fast becoming the shared foundation for robot learning. Palatial V1.0 is the first and only automated asset generator for Newton.
It also generates deformables. Cables that coil and clothing that folds are the content robots struggle with most and simulation has historically lacked. A deformable requires more than geometry: solver choice, timestep, material settings, and contact parameters all have to work together. Palatial authors those settings automatically. Each asset arrives as a .newton.usda file with the solver, timestep, and tuned parameters already written. You load it, you don't configure it.
Dual-arm cable insertion demonstrates deformable behavior inside a manipulation task.
palatial-newton is open source
We're also open-sourcing palatial-newton at github.com/PalatialSim/palatial-newton, our integration layer for Newton. The public API is one call:
from newton.palatial import load
bundle = load("asset.newton.usda")
# bundle: model, solver, states, control, ready to stepOne asset, three simulators
load() detects whether an asset is rigid, articulated, cloth, or rod. It builds the Newton model and selects the right solver from the parameters in the file. The repo includes runnable examples and real assets: a door, shirt, cable, and bed.
Everything generated by Palatial works in Isaac Sim, MuJoCo, and Newton. Assets are validated against NVIDIA's SimReady Foundation validation suite, with dedicated scenarios for deformables such as cloth drop-and-drape checks. You receive evidence that the asset runs, not a file that merely loads.
V1.0 changes the input from simulation expertise to the data you already have. A datasheet can become a jointed appliance. A photo can become a warehouse container with correct collisions. A CAD file can become a validated Newton simulation.
The pipeline now runs 24/7. The first 10,000 assets were the proving run. The goal is 1 million this year.