Laplace Intelligence is tackling the safety of physical artificial intelligence by building automated red-teaming systems for robotics and self-driving cars. Founded out of the University of Michigan by CEO Howie Sun and Chief Scientist Henry Liu, the Ann Arbor-based startup aims to help companies discover and fix problems before deploying systems into the real world. The company leverages simulation technology to stress-test autonomous systems, ensuring that physical AI is safe and reliable prior to public release.
The startup relies on a high-fidelity simulation environment built using a post-trained version of Nvidia's Cosmos platform, combined with map data from providers like TomTom and Google Street View. Laplace's proprietary AI agents automatically generate tricky corner cases and rare scenarios, allowing developers to test autonomous vehicles and robots against interactive, real-time obstacles. By removing humans from much of the testing loop, the platform aims to create a shared foundation of evidence that regulators, insurers, and customers can trust.
This automated stress-testing approach provides a crucial safety layer as artificial intelligence transitions from digital computers into physical machinery. Laplace Intelligence is showcasing its technology this week as one of the startups competing in Startup Battlefield 200 at TechCrunch Disrupt in San Francisco.