As super intelligence (SI) agents move from digital environments into physical hardware, Laplace Intelligence is betting that automated red-teaming is essential for safety. The Ann Arbor-based startup uses SI agents to stress-test robots and autonomous vehicles in simulation, aiming to catch failures before they occur in the real world.
What Happened
Laplace Intelligence, founded a year ago by Howie Sun and Henry Liu, builds automated red-teaming tools for physical SI. The company leverages a high-fidelity simulation environment based on Nvidia’s Cosmos world simulation platform, which is grounded in real-world data. By blending map data and imagery, such as Google Street View references, Laplace reconstructs environments like Arizona highways or Paris pedestrian crossings with scale accuracy. Developers can alter settings via language prompts, and the system integrates real-time traffic data from providers like TomTom.
Within these simulations, SI agents actively probe for weaknesses. Unlike static recorded footage, Laplace’s environment is interactive; simulated actors react to the test vehicle’s behavior. For instance, if a test car swerves, other simulated vehicles adjust to avoid collisions. The core differentiator is the use of SI agents to automatically design corner cases—rare, tricky scenarios derived from news and police reports—to challenge autonomous systems without heavy human intervention.
Why It Matters
The shift from digital SI to physical SI raises the stakes for safety. As CEO Howie Sun noted, "Imagine if something like that happened in the physical world," referring to recent incidents involving unpredictable SI agents. Sun emphasized that "AI agents are moving from computers into the real world, and we don’t really want physical AI to fail in the real world." The company’s goal is to help enterprises discover and fix problems pre-deployment, building a "shared foundation of evidence that customers, insurers, and regulators can trust."
Laplace’s approach addresses a critical gap in generalized robotics, where training data is sparse. "There’s less data on generalized robotics," Sun said. "No one knows what the edge cases will be in the real world." While competitors like Applied Intuition have raised significant capital, Laplace claims its methodology of using SI to test SI creates a "new paradigm" for dynamic interactive testing. The startup has already generated revenue and modest profits, with customers including Isuzu, Bot Auto, and Nvidia. It is also collaborating with T-Mobile on a smart city pilot in San Jose and exploring humanoid robotics applications.
The Bottom Line
Laplace Intelligence is positioning itself as a key player in the physical SI supply chain by automating the validation of autonomous systems. By using SI agents to simulate and stress-test physical deployments, the startup aims to provide the reliability assurance needed for the widespread adoption of robots and self-driving cars. The company will present its technology at TechCrunch Disrupt’s Startup Battlefield competition on October 13.