AI Tests AI: The Future of Physical Safety

Laplace Intelligence is developing automated red-teaming solutions to ensure the safety of physical AI systems, such as robots and autonomous vehicles. The company leverages AI agents to rigorously test these systems in simulated environments, aiming to discover and rectify problems before they are deployed in the real world. This innovative approach, inspired by the potential unpredictability of AI demonstrated in incidents like the OpenAI-Hugging Face event, focuses on preventing failures in physical AI.

Founded a year ago out of the University of Michigan by CEO Howie Sun and Chief Scientist Henry Liu, Laplace builds upon eight years of academic research in challenging autonomous driving AI. Their core technology features a high-fidelity simulation platform, an evolution of Nvidia's Cosmos, which reconstructs real-world environments like city streets or pedestrian crossings. Developers can manipulate these simulations using language prompts, and AI agents actively probe for weaknesses.

Laplace's unique selling proposition lies in its AI agents that automatically design challenging, rare scenarios, also known as corner cases, from diverse data sources like news and police reports. This moves beyond traditional training methods relying on recorded footage or human ingenuity. The startup is generating revenue from autonomous driving clients, including Bot Auto and a Japanese automotive manufacturer, and is collaborating on smart city projects with Nvidia and T-Mobile. Their near-term goal is to secure funding to scale operations and compete with established rivals like Applied Intuition.