Laplace Intelligence, founded in Ann Arbor by Howie Sun and Henry Liu about a year ago out of the University of Michigan, is building automated red-teaming for robotics and other physical AI systems. The company argues that as AI agents move from computers into the physical world, unpredictable behavior becomes riskier and that automated, AI-driven testing is necessary to reduce failures before real-world deployment. Sun and Liu’s 2023 paper on challenging autonomous-driving AI appeared on the cover of Nature, which helped generate industry interest.
The company’s approach
Laplace combines a high-fidelity simulation reconstructed from real-world data with AI agents that actively probe for weaknesses and failure modes. The goal is to ensure robots, autonomous vehicles, or smart-city systems are tested and made safe in simulation before being deployed in physical environments.
Technical foundation
The startup uses a post-trained version of Nvidia’s Cosmos platform as the simulation backbone, enabling reconstructions of real locations — from an Arizona highway to a Paris pedestrian crossing — by blending meshes of map data and images such as Google Street View. Rather than producing random scenes, Laplace’s simulator recreates environments to scale and can incorporate live traffic and congestion feeds from navigation providers like TomTom. Developers can also change conditions in real time through language prompts (for example, adding afternoon traffic).
The simulation models not only the environment but also the behavior of other actors. In demos, when a test agent directed a vehicle to swerve into oncoming traffic, the would-be struck car moved to avoid collision instead of behaving unrealistically. Sun describes the platform as an interactive world model in which participants react to one another.
Automated corner-case generation
Beyond the simulated environment itself, Laplace’s key differentiator is the AI agents that automatically design corner cases — rare, difficult scenarios that can cause systems to fail. These agents draw on diverse inputs such as news and police collision reports to craft challenging situations for autonomous vehicles and other physical AI agents, reducing the need for humans to manually invent test cases.
Customers, partnerships and business model
Laplace has begun generating revenue and has achieved modest profitability, with most early business coming from autonomous-driving use cases. Customers include Japanese truck maker Isuzu and Texas-based autonomous trucking firm Bot Auto. Nvidia is also a customer. Laplace, Nvidia and T-Mobile are collaborating on a free smart-city pilot for the City of San Jose to provide high-level traffic operations, and the company is in early-stage partnerships with teams building humanoid warehouse robots.
Most current contracts are project-based; Laplace is working to convert that into recurring revenue and scale the business. Part of its near-term plan is to raise funds and increase scale, a strategy it hopes to advance by competing in TechCrunch Disrupt’s Startup Battlefield. Laplace will present on the Startup Battlefield stage on Monday afternoon, October 13, 2024.
Competition and Laplace’s position
A notable rival is Applied Intuition, which has raised roughly $1.2 billion and holds a reported $15 billion post-money valuation; Isuzu is also listed among Applied Intuition’s customers. Howie Sun believes Laplace’s methodology — using AI to test AI — represents a new paradigm that turns physical-AI testing into a dynamic, interactive process and will serve as the company’s primary moat.
Why this matters
As AI agents leave simulation and operate in the real world, failures carry greater risk. Laplace’s approach of automated, interactive simulations with AI-generated corner cases aims to expose and fix weaknesses before deployment, improving safety for autonomous vehicles, robots and smart-city systems. Scaling that capability will be important as generalized robots and other physical AI systems become more common.
Key facts
- Founders: Howie Sun (CEO) and Henry Liu (chief scientist)
- Origin: founded about one year ago with University of Michigan research background
- Notable milestone: 2023 paper featured on the cover of Nature
- TechCrunch Disrupt Startup Battlefield presentation: Monday afternoon, October 13, 2024
- Customers/partners: Isuzu, Bot Auto, Nvidia, T-Mobile, San Jose smart city pilot
- Main competitor: Applied Intuition (about $1.2 billion raised, $15 billion post-money valuation)
Laplace’s stated goal is to make physical AI safer before real-world deployment by automating red-teaming through high-fidelity, interactive simulations and AI-driven scenario generation.
