Robots and self-driving vehicles can encounter situations that are difficult to anticipate before deployment. Laplace Intelligence is building a way to test those systems in simulation, using AI agents to create challenging scenarios and expose potential weaknesses.
From academic research to a testing product
Laplace was founded a year ago by CEO and co-founder Howie Sun and co-founder and chief scientist Henry Liu, growing out of the University of Michigan. Sun had spent eight years in academia researching how to challenge autonomous driving AI in simulation.
In 2023, Sun and Liu published a paper on the subject that appeared on the cover of Nature. Autonomous vehicle companies then began contacting them about putting the ideas to work. Sun said he wanted to take the research beyond academia and turn it into a useful product.
The company takes its name from “Laplace’s demon,” an 1814 thought experiment by French scholar Pierre-Simon Laplace. It imagined that complete knowledge of the universe’s state could make it possible to predict the future with total precision. Sun describes the startup’s approach as using simulation to predict what may happen in the real world.
A simulated world that reacts
Laplace combines a high-fidelity simulation reconstructed from real-world information with an AI agent that searches for failure points. Its goal is to let developers check physical AI systems—including robots, autonomous vehicles, and smart city solutions—before deployment.
The simulation environment is a post-trained version of Nvidia’s Cosmos, an AI world simulation platform. Laplace says it can recreate settings ranging from an Arizona highway to a Paris pedestrian crossing. A Google Street View image can inform a generated street, while real-world map data helps keep the environment true to scale.
Developers can use language prompts to alter conditions in real time, such as adding afternoon traffic. Sun says Laplace also draws real-time traffic and congestion information from navigation providers like TomTom.
In the simulation, other participants behave as well as the environment. During a demonstration, Sun directed a test agent vehicle to swerve into oncoming traffic. The car in its path moved to avoid a collision. The company says this interactivity distinguishes its simulated world from systems where generated vehicles may fail to respond realistically.
Finding the unusual cases
Laplace’s central pitch is its AI agents, designed to automatically come up with “corner cases”: rare or tricky scenarios that could cause an AI system to fail. These agents can draw on sources such as news and police reports on collisions to challenge autonomous vehicles and other physical AI systems.
Today, autonomous vehicle developers commonly use recorded footage to build simulations of edge cases, or ask people to devise and construct new scenarios. Some companies have already begun using AI to create edge cases for training; the source cites self-driving truck company Waabi as an example.
Sun sees a larger need ahead as generalized robots move into real-world settings. He said there is less data on generalized robotics and that no one knows what its edge cases will be. Testing with AI agents could help developers explore situations they have not already encountered in recorded data.
Customers, competition, and the next step
Laplace has begun earning revenue and has turned a modest profit, despite its founders’ academic backgrounds. Most of its current business comes from autonomous driving projects. Customers include Japanese trucking manufacturer Isuzu and Texas-based autonomous trucking firm Bot Auto. Sun says Nvidia is also a customer.
Nvidia, T-Mobile, and Laplace are working together on a free smart city pilot for the City of San Jose, focused on high-level traffic operations. The startup is also in the early stages of work with partners building humanoid robots for warehouses.
Most contracts are project-based for now, and Laplace is working to develop recurring revenue. It hopes to raise funds and scale, with competing at TechCrunch Disrupt among its near-term ambitions. The company is one of 20 startups in Startup Battlefield, the event’s three-day pitch competition in downtown San Francisco, and is scheduled to present Monday afternoon, October 13.
Laplace faces Applied Intuition, a late-stage startup that has raised $1.2 billion at a $15 billion post-money valuation; Isuzu is also one of Applied Intuition’s customers. Sun argues that Laplace’s method of using AI to test AI could set it apart by making physical AI testing dynamic and interactive. The company’s success will depend on turning that approach into a business that can grow beyond individual projects.