Build physically-accurate worlds
SimReady assets carry measured mass, friction and restitution. A low-μ object actually slips the gripper.
Calibrated worlds where simulated data is worth as much as real. SimReady assets, humanoids and arms, synthetic data and a robot-agnostic benchmark — running entirely on WebGL, in a tab.
One robot station, swappable worlds, identical calibrated physics. A policy — or a benchmark — moves between them unchanged.
Kitchen, warehouse, lab, living room — swappable scenes that share one calibrated robot station.
Browse environments →GR-1 & Atlas-style humanoids plus Panda / UR / industrial arms — one IK, one benchmark.
Meet the robots →Tableware, produce, tools, lab glass, deformables — each measured, tagged, physically grounded.
Open the library →One demonstration becomes thousands of labeled episodes — then every robot is scored under identical criteria.
SimReady assets carry measured mass, friction and restitution. A low-μ object actually slips the gripper.
Teleoperate or autopilot once in the calibrated sim, then multiply it with provenance-banded domain randomization.
Every robot attempts every task under identical success criteria — RoboFinals-style, humanoid or arm.
The browser sells the demo; the same task & schema scale to thousands of GPU-parallel environments.
Pick a robot, run a demonstration, multiply it into a dataset — live, in your browser.
Launch the simulator →