Michigan spinout Laplace Intelligence pitches AI red-teaming for robots on a tuned Nvidia Cosmos simulator

Laplace Intelligence, an Ann Arbor company spun out of the University of Michigan about a year ago, is building automated red-teaming for robots and autonomous vehicles, TechCrunch reported on Oct 13 ahead of the start-up's Startup Battlefield pitch at TechCrunch Disrupt.

Chief executive Howie Sun and chief scientist Henry Liu, his former PhD adviser, base the product on their 2023 Nature cover paper on adversarial testing of driving AI: a post-trained version of Nvidia's Cosmos world model rebuilds real streets from map and imagery data, and AI agents generate rare corner cases, including scenarios drawn from news and police collision reports, to make the system under test fail before deployment.

The company did not disclose funding or customers. MEKAWIRE has covered one direct peer this month: SafeWorld, which came out of stealth with a $12.2 million seed round on Oct 5 to sell safety simulation for generative-AI robots, a sign that pre-deployment testing is turning into its own funded category.