AI agents put physical AI systems through their paces
Before robots and autonomous vehicles operate in everyday settings, Laplace Intelligence wants to test how they respond to the unexpected. The Ann Arbor startup is developing an automated system that uses AI agents to challenge physical AI in detailed simulations and help developers identify issues before deployment.
Co-founded by CEO Howie Sun and chief scientist Henry Liu, Laplace grew out of research at the University of Michigan. Their work on testing autonomous-driving AI was published in Nature in 2023, and the pair later built the research into a commercial product.
The platform recreates real-world settings, from highways to city crossings, using map information and other data. Developers can adjust conditions with prompts, such as adding traffic, while simulated road users respond to one another. AI agents then generate challenging, less-common scenarios--often called edge cases--to assess how a system behaves.
Laplace says its technology is being used in autonomous-driving projects, with customers including Isuzu and Bot Auto. The company also works with Nvidia and T-Mobile on a smart-city pilot in San Jose and is exploring applications in warehouse robotics.
As robots and self-driving systems take on more tasks, automated simulation could give developers a repeatable way to examine their behavior before real-world use. Such testing may become an important part of building confidence in increasingly capable physical AI.
