The recent OpenAI-Hugging Face incident underscored a crucial point: AI agents can behave unpredictably.
“Imagine if something like that happened in the physical world,” Howie Sun, CEO and co-founder of Laplace Intelligence, shared with TechCrunch. “AI agents are moving from computers into the real world, and we don’t really want physical AI to fail in the real world.”
Laplace, based in Ann Arbor, is dedicated to developing automated red-teaming solutions—a rigorous stress-testing method designed to intentionally expose system vulnerabilities—specifically for robotics and physical AI. The company operates on the premise that only AI can adequately test AI to ensure the safety of physical systems.
“Our goal is to help companies discover and fix problems before deployment — and, over time, build a shared foundation of evidence that customers, insurers, and regulators can trust,” Sun stated.
Laplace is among the 20 promising startups competing in Startup Battlefield 200, an innovation pitch competition scheduled for TechCrunch Disrupt in San Francisco from October 13-15 this week. The company is set to present on the Disrupt Stage on Tuesday afternoon, October 13.
Sun, along with his co-founder and chief scientist Henry Liu, launched Laplace one year ago, originating from the University of Michigan. Sun dedicated eight years to academic research, focusing on methods to challenge autonomous driving AI in simulated environments. In 2023, he and Liu—his PhD advisor at the time—published a paper on their findings, which was featured on the cover of the prestigious scientific journal Nature. This publication attracted significant interest from autonomous vehicle companies keen on applying their concepts.
“I realized that I wanted to build this and take it from purely academic research into a product that’s useful in the real world,” Sun reflected.
Building upon their academic foundation, Sun and Liu established Laplace, named after “Laplace’s demon,” an 1814 philosophical concept by French scholar Pierre-Simon Laplace, who theorized that perfect knowledge of the universe’s current state would allow complete prediction of its future.
“We are essentially using simulation to predict the real world’s future,” Sun explained.
At the heart of the startup’s offering is a high-fidelity simulation, meticulously recreated from real-world data, coupled with an AI agent specifically designed to continually identify vulnerabilities and failure points. The primary objective is to ensure that any physical AI—whether a robot, an autonomous vehicle, or a smart city solution—achieves a high level of safety in simulation before its actual deployment.
The simulation environment leverages a post-trained version of Nvidia’s Cosmos, an AI world simulation platform, which is rooted in real-world data, enabling Laplace to accurately recreate diverse settings, from an Arizona highway to a pedestrian crossing in Paris. Simulations are generated by seamlessly integrating various real-world data sources. For instance, a Google Street View image can serve as a basis for generating an entire street. Rather than creating random scenarios, the simulator intelligently combines actual map data to reconstruct environments true to scale.
Developers can then use natural language prompts to modify the setting dynamically, such as introducing afternoon traffic. Sun noted that Laplace integrates real-time traffic and congestion data from navigation providers like TomTom.
Within Laplace’s simulations, not only is the environment simulated, but the behavior of other entities is also emulated. For example, during a demonstration, Sun instructed the test agent vehicle to swerve into oncoming traffic. The vehicle that would have been hit—or driven through, as is common in some generative world models—dynamically moved to avoid a collision.
“It’s truly an interactive simulation environment,” Sun emphasized. “Everyone is reacting to everything … This is more like a world model with interactive participants.”
While the simulated environment is a key component of Laplace’s product, its true differentiating factor lies in the AI agents the startup is developing to autonomously design “corner cases”—complex, rare scenarios that can confound AI systems—thereby further minimizing human intervention.
“They can design from the news, from police reports on collisions — all kinds of information to actively challenge the autonomous vehicles and other physical AI agents,” Sun explained.
Currently, autonomous vehicle developers typically train AI to manage edge cases using recorded footage converted into simulations, or by relying on human input for new scenario ideas and designs. While many AV companies, such as self-driving truck firm Waabi, have begun incorporating AI for generating training edge cases—mirroring Laplace’s approach—the critical need for this technology will become even more pronounced with the deployment of generalized robots in real-world settings.
“There’s less data on generalized robotics,” Sun noted. “No one knows what the edge cases will be in the real world.”
Despite Sun and Liu’s academic backgrounds and lack of traditional Silicon Valley tech pedigrees, the duo has successfully generated revenue and achieved a modest profit. A significant portion of this income originates from autonomous driving applications, with clients including Texas-based autonomous trucking company Bot Auto and a major Japanese automotive manufacturer.
Laplace is also collaborating with Nvidia and T-Mobile on a complimentary smart city pilot project for San Jose, California, aimed at enhancing high-level traffic operations. Additionally, Laplace is in the early stages of partnerships with companies developing humanoid robots for warehouse environments.
Presently, most of Laplace’s customer agreements are project-based, and the company is actively working towards converting these into recurring revenue streams. The startup’s immediate objective—which it hopes to advance by competing at Disrupt—is to secure funding and scale its operations to effectively compete with rivals.
A primary competitor is Applied Intuition, a prominent late-stage startup that has accumulated $1.2 billion in funding to date, achieving a $15 billion post-money valuation.
Sun remains unconcerned, believing that Laplace’s fundamental approach of using AI to test AI represents a “new paradigm that completely transforms the testing of physical AI,” evolving it into a “dynamic interactive testing process.”
“That will be the biggest moat,” he concluded.
This article has been updated to clarify Laplace’s relationship with Nvidia.