Laplace Intelligence conducts red-team testing on robots and self-driving cars before they enter the real world.

Meta description: Laplace Intelligence uses AI red teaming and Nvidia Cosmos simulations to test robotics and autonomous vehicles before real-world deployment.

Tags: AI red teaming, Laplace Intelligence, Nvidia Cosmos, physical AI, autonomous vehicles

Featured image: abstract simulation grid with glowing road networks, vehicle paths, and sensor overlays; no real people, no company logos, no text.

After the OpenAI-Hugging Face incident showed that AI agents can be unpredictable, Laplace Intelligence is arguing that physical AI needs stricter safety checks before leaving the computer. The company will present on the Startup Battlefield stage at TechCrunch Disrupt on Monday afternoon, October 13, in downtown San Francisco. It is one of 20 startups competing in the three-day pitch competition. Its approach is automated AI red teaming: use AI to stress-test robots, autonomous vehicles, and smart city deployments before they operate in the real world. Laplace bets that the only way to make physical AI safe is to have AI test it.

Background: From Nature Research to a Physical AI Testbed

Laplace is based in Ann Arbor and was founded about a year ago out of the University of Michigan by CEO and co-founder Howie Sun and co-founder and chief scientist Henry Liu. Sun had spent eight years in academia researching how to challenge autonomous driving AI in simulation. In 2023, Sun and Liu, his then doctoral advisor, published a paper on that subject that made the cover of Nature. The publication brought calls from autonomous vehicle companies, pushing Sun to build a product.

The startup's name comes from Laplace's demon, an 1814 thought experiment by French scholar Pierre-Simon Laplace: if one knew the exact state of everything in the universe, one could predict the future with total precision. Sun says Laplace is essentially using simulation to predict the real world's future. Sun told TechCrunch that if something like the OpenAI-Hugging Face incident happened in the physical world, the consequences would matter more, because AI agents are moving from computers into real systems and people do not want physical AI to fail in the real world.

Development: Simulating Streets and Designing Corner Cases

At the core of the product is a high-fidelity simulation reconstructed from the real world, paired with an AI agent that constantly probes for weaknesses and failure points. Laplace says the goal is to help companies discover and fix problems before deployment and build shared evidence that customers, insurers, and regulators can trust.

The simulation environment is a post-trained version of Nvidia Cosmos, an AI world simulation platform grounded in the real world. Laplace can reconstruct anything from an Arizona highway to a Paris pedestrian crossing. Simulations blend different real-world information rather than generate random scenes. A Google Street View image can seed a whole street, while real-world map data keeps it true to scale. Developers can use language prompts to change settings in real time, such as adding afternoon traffic. Sun says Laplace pulls real-time traffic and congestion information from navigation providers like TomTom.

The simulation also models other actors. In a demo, Sun directed a test agent vehicle to swerve into oncoming traffic. The car it would have hit moved out of the way to avoid a collision. Sun called it a truly interactive world model with participants reacting to one another.

The startup says its unique selling point is AI agents that create corner cases, tricky rare scenarios that could trip up an AI system. The agents can draw from news, police reports on collisions, and other information to actively challenge autonomous vehicles and other physical AI agents, moving the human further out of the loop. Today, many autonomous vehicle developers train AI to handle edge cases using recorded footage that they then turn into simulations, or by relying on humans to design new scenarios. Some, including self-driving truck company Waabi, have started using AI to create edge cases for training. Sun says the need becomes most apparent when generalized robots enter the real world, because there is less data on generalized robotics and no one knows what the edge cases will be.

Latest Progress: Pilots, Profit, and Startup Battlefield

Laplace says it has started generating revenue and even a modest profit, despite the founders' academic, rather than Silicon Valley, backgrounds. Most of the money comes from autonomous driving use cases. Customers include Japanese trucking manufacturer Isuzu and Texas-based autonomous trucking firm Bot Auto. Nvidia is also a customer, Sun said.

Laplace, Nvidia, and T-Mobile are working together on a free smart city pilot for the City of San Jose to provide high-level traffic operations. The company is also in the early stages of working with partners building humanoid robotics for warehouses. Most of Laplace's customer contracts are project-based today, and the company is working toward recurring revenue. Its near-term goal, helped by Disrupt visibility, is to raise funds and scale to compete with rivals.

One prominent rival is Applied Intuition, a late-stage startup that has raised $1.2 billion to date at a $15 billion post-money valuation. Isuzu is also one of Applied Intuition's customers. Sun said he is not too worried, because Laplace's core method of using AI to test AI is a new paradigm that changes the testing of physical AI into a dynamic interactive testing process. He called that shift the biggest moat.

Laplace now sits at the transition point between academic research, early commercial pilots, and a pitch competition that could bring more visibility. The next test is whether project-based revenue can become recurring revenue, and whether automated red teaming can become part of the evidence base for physical AI before robots and vehicles are deployed at scale.