Laplace Intelligence tests robots and self-driving cars in simulation

Ann Arbor-based Laplace Intelligence uses AI agents and interactive simulations to test robots and self-driving vehicles for potential failures before deployment. The University of Michigan spinout calls the process automated red-teaming.

CEO and co-founder Howie Sun told TechCrunch that the company aims to help customers find and fix problems before deployment, then build evidence that customers, insurers and regulators can trust. Sun founded Laplace a year ago with co-founder and chief scientist Henry Liu.

The company’s simulation platform is based on a post-trained version of Nvidia’s Cosmos. Laplace says it can reconstruct environments using real-world information, including map data and Google Street View images, and adjust settings with language prompts. In a demonstration described by TechCrunch, a simulated car moved to avoid a test vehicle that Sun directed into oncoming traffic.

Laplace is developing AI agents to automatically create rare, difficult scenarios, or corner cases, to challenge physical AI systems. The company says current autonomous-vehicle testing often uses recorded footage or scenarios designed by people.

Laplace has generated revenue and a modest profit, according to TechCrunch. Sun named Isuzu, Bot Auto and Nvidia as customers. Laplace, Nvidia and T-Mobile are also working on a free smart city pilot for San Jose. The startup is one of 20 companies competing in TechCrunch Disrupt’s Startup Battlefield, and hopes the event will help it raise funds and scale.