Founded by CEO Howie Sun and chief scientist Henry Liu, Laplace emerged from University of Michigan research that previously earned a cover feature in Nature. The company’s platform relies on a post-trained version of Nvidia’s Cosmos, allowing developers to reconstruct precise real-world environments—from Arizona highways to city intersections—using map data and street-level imagery. Unlike traditional static simulators, these environments feature interactive participants that react dynamically to test scenarios.
Laplace Intelligence bets on AI-driven stress testing for robotics
As autonomous systems migrate from controlled labs to unpredictable public spaces, Ann Arbor-based Laplace Intelligence is automating the red-teaming process. By using AI to probe for failure points in high-fidelity simulations, the startup aims to ensure self-driving vehicles and humanoid robots survive the real world before deployment.
Laplace differentiates itself by moving humans out of the loop, employing AI agents to synthesize edge cases from police reports and news data. This approach addresses a critical data gap in the emerging field of generalized robotics, where failure modes remain largely unmapped. The startup has already secured revenue from industry players including Isuzu, Bot Auto, and Nvidia, while collaborating on a smart city pilot in San Jose. Despite facing stiff competition from industry giants like Applied Intuition, Sun maintains that their methodology of AI-driven, interactive testing provides a distinct competitive advantage in a market where safety standards are still being defined.
Comments (0)
No comments yet. Be the first!