Evaluate any robotics simulation
on any engine.
Sim2Sim converts a full simulation experiment between engines — assets, contact physics, control scripts, and the runtime environment. Drop a project, get it running anywhere.
Drop your simulation here
.zip · .usd · .urdf · .mjcf · python project
- Isaac Sim
- MuJoCo
- Drake
- PyBullet
- Gazebo
- Genesis AI
...and more.
Sim2Sim versus the best open-source converter.
Same manipulation policy, same seed. Here's what happens when you port it to Isaac Sim with the leading open-source pipeline — and with Sim2Sim.
Authored experiment in MuJoCo. Policy reaches the target and completes the task.
Geometry imports into Isaac Sim, but contacts drift and the policy fails to complete the task.
Assets, contacts, control script, and runtime — converted. Policy succeeds at parity in Isaac Sim.
Not just an asset converter.
Sim2Sim translates the whole experiment — assets, physics, control scripts, and runtime environment — so a policy that works in one engine actually runs in another.
- 01Deterministic
Asset conversion
Geometry, kinematic chains, meshes, materials. Structural diff verifies joint count, mesh count, and the kinematic tree match the source.
- 02AI-assisted + optimization
Simulation conversion
Map contact parameters, actuator gains, and solver settings between engines. Open-loop physics baselines establish residual error.
- 03Hybrid AST + AI
Script translation
Parse source code, transform API calls and control flow. Per-component verification on recorded states for observation, reward, reset, action.
- 04Deterministic
Environment capture
Detect the source environment and emit a target spec — Dockerfile, CUDA, target SDK. The converted model loads and runs end-to-end.