# Laplace Intelligence Builds AI Simulations to Test Autonomous Systems

By Simon Yoon

Canonical URL: https://www.tokenpost.com/news/technology/31392
Published: 2026-10-13T17:36:34.000Z
Updated: 2026-10-13T17:36:34.000Z
Section: Technology

> The University of Michigan-founded company models rare and dangerous scenarios for vehicles, robots, cities, warehouses and airports before deployment.

Laplace Intelligence is developing AI-driven simulations that expose autonomous vehicles, robots and other physical-AI systems to rare or dangerous scenarios before deployment, aiming to improve safety testing.

Laplace was founded out of the University of Michigan by CEO Howie Sun and chief scientist Henry Liu. Sun described the company’s work as part of a shift in artificial intelligence as systems move beyond software environments.

“AI agents are moving from computers into the real world,” Sun said.

Laplace describes its platform as an “agentic world model” combining 3D-world reconstruction, multi-agent behavior modeling and counterfactual simulation. The system is positioned for autonomy programs, cities, warehouses and airports.

The platform can use multi-camera imagery, satellite data, LiDAR and high-definition maps to create editable 3D environments. Developers can then model interactions among multiple agents and test alternate outcomes in simulation.

The company’s research foundation includes a 2023 University of Michigan study published in Nature. Researchers used AI-controlled background agents trained to perform adversarial maneuvers in autonomous-vehicle simulations. The method accelerated vehicle-safety evaluation by 1,000 to 100,000 times in the study’s experiments.

“Everyone is reacting to everything,” Sun said, describing the interactions modeled in the simulations.

A separate 2023 study by University of Michigan researchers, published in Nature Communications, introduced NeuralNDE. The simulator was designed to reproduce normal and safety-critical driving statistics using real-world data.

That research used vehicle trajectories collected at 2.5 hertz from a two-lane roundabout at State Street and West Ellsworth Road in Ann Arbor, Michigan. The data was used to model driving conditions that included both ordinary and safety-critical situations.

Laplace’s broader positioning reflects the challenge of testing autonomous systems against events that occur infrequently in the real world. Simulation allows developers to create and vary interactions in controlled environments while evaluating how autonomous systems respond.