Prompt to Simulated World. Train Robots at Scale.
Generate physics-accurate 3D environments from text.
Native to NVIDIA Isaac Sim.
isaac-worlds generate
Prompt
"Warehouse, metal shelves, concrete floor, overhead lighting, scattered cardboard boxes"
Parse semantics
Layout generation
Asset placement
PhysX properties
USD composition
OpenUSD|PhysX 5
847
Objects
32
Materials
128
Joints
512
PhysX
NVIDIA Ecosystem
Isaac Sim
Isaac Lab
Omniverse
OpenUSD
PhysX
ROS 2
Weeks to Minutes
Manual Pipeline
3D Scene
weeks
Physics
weeks
Assets
weeks
Validation
weeks
Weeks
Isaac Worlds
3D Scene
minutes
Physics
minutes
Assets
minutes
Validation
seconds
Minutes
Orders of magnitude
faster
Fraction of
the cost
Unlimited
variants
Three Steps
Describe
Natural language prompt
prompt.txt
1scene: warehouse_interior
2floor: concrete, worn
3objects: [shelves, boxes]
4lighting: overhead, variable
5physics: rigid_body
Generate
Physics-accurate 3D world
Scene Builder
Building
OpenUSD
PhysX|Rigid Body
Train
RL, IL, or synthetic data
Training Monitor
Epoch 847
94.2%
Success
847.3
Avg Return
12K fps
Sim Speed
Built for Robot Training
Fast
generation
Prompt-to-World
0
conversion steps
Isaac Sim Native
PhysX
GPU-accelerated
Physics-Accurate
Infinite
unique variants
Domain Randomization
1:1
compatibility
OpenUSD Pipeline
OSMO
orchestration
Cloud Scale
The NVIDIA Stack
Isaac Worlds
Prompt-to-World Engine
Isaac Lab
Robot Learning
Isaac Sim / Omniverse
Simulation & Rendering
PhysX / OpenUSD / Cosmos
Physics, Assets, World Models
NVIDIA GPU / DGX / OVX / OSMO
Compute Infrastructure
Training Scenarios
Terrain: Varied
IMU Active
Locomotion
RLWhole-Body Control
6-DOF Arm
Contact: PhysX
Manipulation
Imitation LearningPhysX
Lidar + RGBD
cuVSLAM
Navigation
Visual SLAMPath Planning