Laplace makes simulated traffic react to robots under test
Laplace Intelligence is developing simulations in which the surrounding vehicles react to the autonomous system being tested. Its Michigan research roots now support a product combining street imagery, map data and language-based changes to traffic conditions. The startup is also developing agents that create challenging scenarios from collision reports, with autonomous-driving customers supplying most of its current revenue.
Artificial Intelligence··Night
Other vehicles respond inside the simulation
Laplace Intelligence is developing simulations for testing autonomous vehicles and robots before they are deployed in the physical world. The Ann Arbor company grew out of University of Michigan autonomous-driving research by Howie Sun and Henry Liu. Its system models both the surroundings and the actions of other participants. That lets developers expose a system under test to a changing environment rather than a scene populated by objects that never react.[1], [2]
In a demonstration described by TechCrunch, Sun directed the test vehicle to swerve into oncoming traffic. The vehicle it was approaching moved away to avoid a collision. The example illustrates how other road users participate in the simulation. Sun said the environment allows everyone to react to one another. The company is showing interactions generated inside its system while developing a product intended to challenge physical AI before deployment.[1]
Street images and maps anchor changing scenarios
The simulation uses a post-trained version of Nvidia Cosmos, a platform for modelling physical environments. Street imagery can supply the visual reference, while map data keeps the reconstructed street at the correct scale. Sun also described incorporating traffic and congestion information from navigation providers such as TomTom. Developers can change the setting through language instructions, for example by introducing afternoon traffic into the recreated road environment.[1]
Laplace is also developing agents intended to design rare and difficult situations using collision news and police reports. These agents would supply scenarios for challenging autonomous systems. Sun said there is less information about unusual cases for general-purpose robots than for autonomous driving. The planned work extends the company’s tests beyond changing traffic density, towards creating situations that can expose failures before machines operate around people and physical objects.[1]
Autonomous driving supplies most current revenue
Most of Laplace’s current revenue comes from autonomous-driving projects. Its customers include Isuzu and Texas-based autonomous trucking company Bot Auto; Sun also named Nvidia as a customer. Laplace is working with Nvidia and T-Mobile on a free smart-city pilot for San Jose. Work with warehouse-humanoid developers is at an early stage. The company’s contracts are mostly project-based, and it is seeking recurring revenue as it expands beyond its research origins.[1]
