Laplace Intelligence Develops Red-Teaming for Physical AI and Autonomous Systems
Laplace Intelligence is developing automated red-teaming solutions to stress-test robotics, autonomous vehicles, and other physical AI systems before deployment. The Ann Arbor-based startup uses high-fidelity simulations grounded in real-world data to identify potential failure points and ensure safety. CEO Howie Sun emphasizes the importance of testing these systems thoroughly given their increasing presence in the real world. Laplace's technology is built on a post-trained version of Nvidia’s Cosmos simulation platform, allowing it to reconstruct environments like highways or pedestrian crossings with high accuracy. The simulations include interactive elements such as traffic and other actors that respond realistically. This approach helps developers uncover rare edge cases before systems are deployed in the real world. The company was founded by Sun and his co-founder Henry Liu after they published influential research on autonomous driving AI stress-testing, which led to interest from major vehicle companies.
Topics: AI safety OpenAI-Hugging Face incident robotics testing startup funding