Ann Arbor-based Laplace Intelligence is pioneering a new approach to ensure physical AI, from robots to self-driving cars, are ready for the real world. By building automated red-teaming in simulation, Laplace aims to discover and fix issues before deployment, using AI to predict and challenge edge cases. This method, rooted in academic research, is now a product that can be applied in the real world, with early customers like Isuzu and Bot Auto already benefiting.

The company's founders, Howie Sun and Henry Liu, have a vision for a safer, more reliable future where AI is constantly tested and improved. They use high-fidelity simulations, like a post-trained version of Nvidia's Cosmos, combined with AI agents that probe for weaknesses and failure points. This ensures that any physical AI is safe in simulation before it's deployed, addressing the unpredictability of AI in the real world.

While competitors like Applied Intuition are making waves, Laplace's approach is rooted in the unique challenges of generalized robotics, where data is scarce and edge cases are unpredictable. With projects in smart cities and warehouses, Laplace is not just testing AI, but shaping its future to be more reliable and secure.

By using real-time data and interactive simulations, Laplace is creating a dynamic testing environment that can adapt to new information. This could be a game-changer in ensuring the safety of physical AI, paving the way for a future where AI is more reliable and less prone to failure.