2 min read
Add as a preferred source on Google

Laplace Pairs NVIDIA OmniDreams With TeraSim for Vehicle Testing

Laplace World creates editable 3D environments from imagery and maps. Earlier research describes related testing methods but did not evaluate the commercial platform.

A test vehicle waits at an empty track intersection at dawn / TokenPost.ai
A test vehicle waits at an empty track intersection at dawn / TokenPost.ai

Laplace Intelligence is pairing NVIDIA's OmniDreams model with simulated traffic behavior to build editable road environments where developers can test autonomous vehicles against difficult situations before deployment.

OmniDreams, now called Cosmos-Dreams, is a world model derived from NVIDIA's Cosmos work. Laplace World pairs its rendering with TeraSim traffic behavior, bringing generated road scenes and modeled activity into one testing environment.

The platform can assemble editable 3D settings from multi-camera imagery, satellite data, LiDAR and high-definition maps. Vehicle trajectories, pedestrian flows and robot routes help model how participants move through those settings.

Users can change a scene and examine how an autonomous system handles a different set of conditions. For vehicle developers, that means rare or hazardous situations can be examined in a controlled setting before deployment.

Laplace's work beyond vehicle testing extends to robots, cities, warehouses and airports. Those settings bring other moving participants into a test, including pedestrians and robots, alongside the autonomous system being evaluated.

The company's approach has roots in two distinct studies published in 2023. Haowei Sun and Henry X. Liu co-authored one that trained an AI tester to focus on dangerous driving situations. The researchers conducted augmented-reality trials with a highly automated vehicle on highway and urban tracks.

The other study introduced NeuralNDE, a simulator designed to model interactions among multiple road users and reproduce statistical patterns in normal and safety-critical traffic. The studies examined research methods; neither was an evaluation of Laplace World's commercial performance.

Laplace CEO and co-founder Howie Sun described a challenge that extends beyond cars. “There’s less data on generalized robotics. No one knows what the edge cases will be in the real world,” Sun said.

Laplace was among 20 companies selected for Startup Battlefield, held Oct. 13-15 in San Francisco. Intero Biosystems won the competition's $100,000 prize.

Simon Yoon

Reporter

Simon Yoon reports on blockchain technology for TokenPost. Send corrections or tips to info@tokenpost.com.

Loading…