Available Now
Book a demo and get early access. Free trial!
Simulation tools have advanced—but most engineering teams still waste time on manual, one-off setups. CAD files arrive in bad shape. Scripts are copied and lost. And if something breaks two weeks later, no one knows what changed.
In an age of GPU-powered physics and AI agents, your simulation prep shouldn't feel like duct tape.
This post shows how to set up a reproducible, scalable simulation pipeline using:
Raw CAD files (STEP, IGES, SOLIDWORKS) are not simulation-ready. They're missing:
Champion helps automate this process further by offering AI-powered CAD import assistance. Once you drop in your STEP or IGES file, Champion not only converts it to USD, but also applies default or inferred physical properties—like mass based on volume, or joint constraints based on part naming patterns. It tags components with semantic labels and suggests hierarchy cleanups, making downstream scripting and simulation setup far easier. This makes Champion ideal for teams that want clean simulation-ready data without spending hours in GUI tools.
Before any simulation can run, a digital representation of the system is needed. This includes geometry, mass/inertia, and connectivity—all of which must be explicitly defined post-CAD import.
Simulation success is built on traceability. Each simulation run typically depends on a mix of CAD inputs, converted USD scenes, Python scripts, and config files like YAML or JSON. Without version control, there's no way to roll back broken changes, compare experimental results across iterations, or share a reproducible setup with your team. That’s why versioning is essential.
Use Git + Git LFS to handle large 3D files efficiently, and consider tools like lakeFS or DVC for more advanced data tracking needs. Include everything—CAD source, converted USD, and all scripts—in one versioned repository for maximum clarity and team alignment.
In Isaac Sim, everything can be automated:
Organize your code into reusable modules:
scene_setup.py
run_episode.py
data_logger.py
The use of simulation scripting enables faster testing cycles and reduced operator intervention. Champion extends this further by introducing Agentic AI for USD scripting. Instead of writing and debugging simulation scripts manually, Champion lets you describe what you want in plain language—like "set up a warehouse robot test with slippery floors and pallet obstacles"—and the AI generates the required USD and Python code. Behind the scenes, every agent in Champion connects to a language model, memory, and a set of APIs. It understands scene graphs, physics constraints, and simulation logic, and can reason through multiple attempts to reach a valid setup.
Once your workflow is clean and repeatable, go big:
Parallel simulation refers to running many instances of a simulation simultaneously, each with different inputs or conditions. A "logical process" (LP) simulates part of the system in sync with others. In practice, this might look like testing 100 variations of a robot's environment—all running at once on GPU nodes—greatly accelerating insight generation and model validation.
Isaac Sim + Omniverse Nucleus make it possible to orchestrate distributed jobs at scale.
Simulation is no longer a sandbox—it's a data factory for AI.
Parallel simulation enables faster runtime and more complex system modeling. Each logical process operates concurrently, bound by local causality constraints.
💡 Champion makes this easy—offering AI-native tools for CAD import, USD scripting, and parallel job execution. Ready to see it in action? Get started with Champion →
Book a demo and get early access. Free trial!