Laplace Intelligence Uses AI to Red-Team Robots and Self-Driving Cars
Laplace Intelligence uses Nvidia Cosmos and AI agents to stress-test robots and self-driving vehicles before deployment for Isuzu, Bot Auto and Nvidia.
Summary
The OpenAI and Hugging Face incident highlighted unpredictable AI agents; Ann Arbor based Laplace Intelligence is automating red-teaming so robots, autonomous vehicles and smart-city systems fail in simulation, not after deployment. CEO Howie Sun and chief scientist Henry Liu founded Laplace a year ago from University of Michigan research. Sun spent eight years studying simulated challenges for driving AI, and his 2023 Nature cover paper with then PhD adviser Liu drew interest from autonomous-vehicle companies. Named for French scholar Pierre-Simon Laplace’s 1814 “Laplace’s demon” thought experiment, Laplace was one of 20 startups in Startup Battlefield’s three-day San Francisco competition, with a Monday afternoon presentation scheduled for October 13.
Laplace post-trains Nvidia’s Cosmos world-simulation platform to reconstruct true-scale locations, including Arizona highways and Paris pedestrian crossings, from sources such as Google Street View, maps and real-time TomTom traffic data. Language prompts modify conditions, simulated actors react interactively, and AI agents generate rare corner cases from news and police collision reports. The approach reduces reliance on recorded footage and human-authored scenarios; Waabi also uses AI-generated cases, but generalized robotics has less data and unknown real-world edge cases.
Laplace says projects for Isuzu, Bot Auto and Nvidia generate revenue and a modest profit. Laplace, Nvidia and T-Mobile are running a free City of San Jose smart-city pilot for high-level traffic operations, while unnamed warehouse-humanoid partnerships are beginning. Most contracts remain project-based, so Laplace seeks funding, scale and recurring revenue. Rival Applied Intuition, which also serves Isuzu, has raised $1.2 billion at a $15 billion post-money valuation; Laplace considers its dynamic AI-testing-AI methodology its moat.
Positives
- Isuzu, Bot Auto and Nvidia already buy Laplace’s simulation and testing work.
- Revenue from autonomous-driving projects has made the one-year-old company modestly profitable.
- Sun and Liu’s 2023 autonomous-driving research appeared on Nature’s cover and attracted commercial interest.
- Nvidia and T-Mobile are joining Laplace in a free City of San Jose smart-city pilot.
- AI agents generate corner cases from news and collision reports, reducing dependence on human-designed tests.
Risks & concerns
- Most Laplace contracts remain project-based, leaving recurring revenue unfinished.
- Applied Intuition has raised $1.2 billion at a $15 billion post-money valuation and shares customer Isuzu with Laplace.
- Generalized robotics has less training data, leaving real-world edge cases largely unknown.
- Warehouse-humanoid partnerships remain early and unnamed, limiting evidence of adoption beyond autonomous driving.
- Physical AI failures could cause real-world harm if simulation misses unpredictable behavior before deployment.