Simulation Infrastructure
for Physical AI
Antim Labs builds structured simulation environments, datasets, and eval suites for robotics teams training and testing embodied agents.
The Problem
Simulation authoring
is the bottleneck.
Robot learning stacks are scaling quickly, but useful simulation environments are still slow to create.
Manual sim authoring
- Weeks per environment
- Limited scenario variation
- Fragile exports
- Hard to reproduce failures
Gizmo workflow
01
INPUT
Point cloud · Floorplan · Prompt · Image
02
SCENE
Generated 3D world
03
STRUCTURE
SCALECOLLIDERSARTICULATIONSEMANTICSEVAL TASKS
04
EXPORT
USD · MJCF · SDF
05
EVAL
PASS / FAIL metrics
Meet Gizmo.
Automated sim authoring for robotics.
Turn prompts, images, floorplans, and point clouds into structured simulation environments — with real-world scale, collisions, articulation, semantics, and export-ready geometry.
Built for modern robotics stacks
- OpenUSD
- Isaac Sim
- MuJoCo / MJCF
- Omniverse
- ROS / Gazebo
- Synthetic sensors
- Eval manifests