NVIDIA has introduced new Cosmos simulation recipes that developers can use to generate training data and test control policies for autonomous vehicles, robots and surgical applications.
Simulation · NVIDIA Cosmos
New Cosmos recipes turn one scenario into many test conditions
NVIDIA's recipes let developers generate synthetic training data and test control policies for autonomous vehicles, robots and surgical applications, keeping scene geometry and motion fixed while changing appearance.
2
Core tasks
Data generation for perception and driving, and policy testing before deployment
3
Target domains
Autonomous vehicles, robots and surgical applications
1
Variable changed at a time
Lighting or weather can change while road layout and vehicle trajectories stay the same
Same scene, different conditions
Representative use case: a daytime driving clip converted to a rainy night scene
Original
Daytime driving
Road layout and vehicle trajectories recorded
→
Generated
Rainy night
Same road layout and same vehicle trajectories
How the workflow runs
1. Start
Driving or robot scenario
→
2. Hold fixed
Geometry and motion
→
3. Vary
Lighting and weather
→
4. Use
Generate training data or test a policy
Why it matters
Controllability isolates the effect of one variable, such as weather or lighting
Expands coverage of rare edge cases that are hard to capture on public roads or in clinical settings
Can shorten development cycles where real-world collection is slow and costly
</
Continue reading The rest of this article is for AI News Blitz readers. Choose an option below to keep reading.
Already purchased? Sign in ✓ Signed in — this article isn’t included in your current plan.Unlocking the full article...