Laplace Intelligence stress-tests physical AI in simulation
Original: Laplace Intelligence red-teams robots and self-driving cars before they hit the real world
Why This Matters
Physical AI safety validation is a gap the AV and robotics industries urgently need to close.
Ann Arbor startup Laplace Intelligence automates red-teaming for robots and self-driving cars using AI-driven simulation, founded a year ago by two University of Michigan researchers whose 2023 Nature paper attracted AV industry interest.
Laplace Intelligence is building an automated red-teaming platform specifically for physical AI — robots, autonomous vehicles, and smart city systems. The idea: let AI deliberately try to break AI before anything rolls out into the real world. CEO Howie Sun and co-founder Henry Liu, formerly PhD advisor and advisee at the University of Michigan, spent years in academia stress-testing autonomous driving systems in simulation. Their 2023 paper on the subject landed on the cover of Nature, which brought autonomous vehicle companies calling.
The platform's core is a high-fidelity simulation environment built on a post-trained version of Nvidia's Cosmos world simulation platform. It can reconstruct real-world locations — an Arizona highway, a Parisian crosswalk — by blending mesh data with AI generation, then deploys an adversarial AI agent that continuously probes for failure points. Sun says the name comes from "Laplace's demon," an 1814 thought experiment imagining perfect predictive knowledge of the universe. "We are essentially using simulation to predict the real world's future," he said.
The company's pitch: discover failures before deployment and build an evidence base that customers, insurers, and regulators can trust. Laplace is competing in TechCrunch Disrupt's Startup Battlefield, presenting October 13 in San Francisco.