Ann Arbor-based Laplace Intelligence is developing automated red-teaming — deliberate stress testing — for robotics and other physical AI systems. The company’s premise is that the safest way to verify physical AI is to have AI agents actively search for weaknesses and failure modes before hardware is deployed into the real world.
Founders, origin and research background
Laplace was founded a year ago by Howie Sun and Henry Liu, building on research from the University of Michigan. Sun has spent the past eight years researching how to challenge autonomous driving AI in simulation. In 2023, Sun and Liu published a paper that appeared on the cover of Nature, and that publication triggered industry interest from autonomous vehicle companies looking to implement the approach commercially.
The company’s name references Pierre-Simon Laplace’s 1814 thought experiment; Sun describes their work as using simulation to predict aspects of the real world.
What the technology does
At the core of Laplace’s product is a high-fidelity simulation reconstructed from real-world data, paired with an AI agent that continuously probes for weaknesses. The aim is to validate that robots, autonomous vehicles, and smart-city systems behave safely in simulation before they are deployed.
The simulation environment is a post-trained version of Nvidia’s Cosmos platform. It blends multiple sources of real-world information — Google Street View imagery, map data and other inputs — to recreate environments at scale, from an Arizona highway to a Paris pedestrian crossing. Rather than producing random scenes, the simulator re-creates true-to-scale locations, and developers can use language prompts to modify conditions in real time (for example, adding afternoon traffic). Laplace also pulls real-time traffic and congestion data from navigation providers such as TomTom.
Inside the simulation, other actors’ behaviors are also modeled. In a demo described by Sun, a test vehicle was directed to swerve into oncoming traffic; the oncoming car in the simulation moved out of the way to avoid a collision, illustrating interactive participant behavior rather than a static world model.
AI red teams and corner-case generation
A core differentiator for Laplace is its AI agents that automatically design corner cases — rare and tricky scenarios that can trip up systems. These agents can derive scenarios from news, police reports and other sources to actively challenge autonomous vehicles and other physical AI.
Today many autonomous vehicle developers use recorded footage turned into simulation or rely on humans to design new edge cases. Laplace’s approach moves more of that work into an automated loop, which Sun says will be increasingly important as generalized robots are deployed in the real world, since there is less existing data on edge cases for generalized robotics.
Customers, revenue and pilots
Despite the founders’ academic backgrounds, Laplace has started generating revenue and even modest profit, largely from autonomous driving use cases. Customers include Texas-based autonomous trucking firm Bot Auto and a major Japanese automotive manufacturer. Laplace is also working with Nvidia and T-Mobile on a free smart-city pilot in San Jose, California, to provide high-level traffic operations. The company is in early-stage partnerships with firms building humanoid warehouse robots as well.
Most of Laplace’s contracts are currently project-based; the startup’s near-term goal is to convert those into recurring revenue and to raise funds to scale the business.
Competition and market position
A notable rival is Applied Intuition, which has raised roughly $1.2 billion to date and is valued at about $15 billion post-money. Sun believes Laplace’s methodology — using AI to test AI — represents a new paradigm that could fundamentally change how physical AI is tested, turning it into a dynamic, interactive process that could become the company’s strongest moat.
Near-term plans
Laplace is competing in Startup Battlefield 200 at TechCrunch Disrupt in San Francisco, taking place October 13–15, where it will present on the Disrupt Stage on the afternoon of Tuesday, October 13. The startup aims to use the exposure to help raise capital and scale its operations.
Note: this article has been updated to clarify Laplace’s relationship with Nvidia.
