Palatial Unveils Asset Generator for Embodied AI Training
The company Palatial has unveiled PhysReady, an automated generator for simulator assets aimed at training embodied AI agents. The tool is designed to address the significant lack of tactile data needed for training physical AGI systems, representing a shift in generative AI applications from text and images toward robotics and physical simulation.
- PhysReady can generate physically realistic 3D assets from text, images, videos, or CAD models, including accurate physics metadata like mass, inertia, and friction. - The platform is designed to support a variety of robotics applications, including industrial automation, field robotics for agriculture and mining, and the training of general-purpose humanoid robots. - A key feature of PhysReady is its ability to create millions of variations of an asset with deterministic domain randomization of its geometry, materials, and physics, which is crucial for robust AI training. - Palatial was a third-place winner in the "Data Engines, Synthetic Data and Simulation" category at the 2025 Nebius Robotics and Physical AI Awards. The judging panel for these awards included representatives from major industry players like NVIDIA and OpenAI. - The company, legally named Palatial Platforms Inc., is building what it describes as the data infrastructure for the age of physical AI, aiming to solve the "real-to-sim" transfer problem. - Job listings for roles such as "Senior ML Engineer" indicate a focus on developing foundation models to convert large amounts of sensor data into high-fidelity simulations. - The company is also hiring roboticists to build physical data collection systems to benchmark and validate the sim-to-real transfer of their generated assets.
