← Cases
ChinaManufacturing2026 – ongoing

Li Auto

Neural reconstruction of driving scenes for ADAS simulation

Li Auto doubled ADAS simulation coverage on equivalent compute, NVIDIA reported

WorkingVendor case studyVendor ↗

Problem

Reconstructing road recordings into simulation scenes constrained Li Auto's driver-assistance testing. Separate camera and lidar reconstruction increased engineering work.

Approach

Li Auto integrated NVIDIA Omniverse NuRec into its internal closed-loop simulation platform. Neural reconstruction generated camera and lidar views, while Li Auto's TokenGS model filled out assets from sparse observations.

Results

2×simulation coverage on equivalent compute, according to NVIDIAVendor case study[1]
approximately 50%lower reconstruction cost per scene, according to NVIDIAVendor case study[1]

Takeaway

The reported gains concerned simulation coverage and reconstruction cost; they did not establish better on-road safety.

Categories

Business function
IT and R&D
Innovation type
Process innovation
Evidence
Vendor case study

Sources

Source summaries

[1]NVIDIA2026

NVIDIA described Li Auto's internal scene-reconstruction system and integration with its simulation platform. It reported twice the simulation coverage on equivalent compute and approximately 50% lower reconstruction cost per scene.

Full details are in the original sources.

Added 1 Nov 2026 · Updated 1 Nov 2026
Open case page ↗