Join NVIDIA at NeurIPS 2024

The NVIDIA DRIVE® team is constantly innovating, developing end-to-end autonomous driving solutions that are transforming the industry.

The Large Spatial Model (LSM) leverages foundation model principles to process unposed RGB images into semantic radiance fields. This approach simultaneously infers geometry, appearance, and semantics within a scene and synthesizes versatile label maps at novel views in a single feed-forward pass.

Inspired by human memory, the 3D Gaussian Mapping (3DGM) framework introduces an end-to-end approach for selective scene retention. This self-supervised, camera-only offline mapping system mimics the human ability to prioritize permanent elements over ephemeral ones, which is crucial for robust perception and localization.

NAVSIM presents a novel simulation-based benchmarking tool for end-to-end driving. By unrolling simplified bird’s eye view abstraction, it efficiently computes metrics like progress and time to collision, aligning closer to closed-loop evaluations than traditional methods.