Laplace Intelligence is building automated red-teaming software to stress-test robotics and self-driving cars in simulation before deployment. Co-founded by CEO Howie Sun and Chief Scientist Henry Liu, the Ann Arbor-based startup emerged from the University of Michigan after the founders published a research paper on challenging autonomous driving AI that was featured on the cover of Nature in 2023. Laplace is presenting at TechCrunch Disrupt in San Francisco this week as part of Startup Battlefield 200, showcasing technology designed to help companies discover and fix problems in physical AI systems.
The startup uses a post-trained version of Nvidia's Cosmos AI world simulation platform, combined with real-world map data from providers like TomTom and Google Street View, to reconstruct environments true to scale. Developers can modify settings in real time using language prompts, while interactive AI agents simulate the dynamic behavior of other actors in the environment. Laplace's proprietary AI agents automatically design rare corner cases to probe for weaknesses, moving humans further out of the testing loop and creating a foundation of safety evidence for regulators and insurers.
This automated simulation testing matters because physical AI systems like autonomous vehicles and robots are moving into real-world environments where failures can have dangerous consequences.