Keep pulling the thread on Jinju Woo.
NVIDIA believes that deploying Level 4 autonomous technology across all operational design domains (ODDs) is much better with LiDAR than without it.
Between 2018 and 2023, both new and legacy auto OEMs in China adapted to a single central computer electrical architecture to remain competitive.
The current generation of Mercedes-Benz vehicles is based on a central computer architecture that will be used in all of their vehicles.
NVIDIA has a limited internal supply of GPUs for compute, requiring weekly prioritization meetings, sometimes involving CEO Jensen Huang, to allocate resources between divisions.
NVIDIA's business model for autonomous driving aims to capture a percentage of the revenue generated from every autonomous mile driven in the future.
NVIDIA collaborates with Tesla and other OEMs that build their own in-car inference chips by providing them with cloud-based training and simulation infrastructure.
NVIDIA DRIVE partners participate in a data sharing program where data collected from different OEM car programs is aggregated to train a common autonomous driving model.
NVIDIA uses a neural reconstruction technology called Neurac to generate synthetic training data by modifying real-world data, such as changing backgrounds or vehicle trajectories.
NVIDIA's current approach for its L2++ autonomous driving system is "mapless," meaning it does not rely on pre-built high-definition maps for navigation.
NVIDIA's autonomous driving software architecture runs two stacks in parallel: an end-to-end model and a "classical stack" that acts as a safety guardrail to verify the model's outputs.
NVIDIA's next-generation autonomous driving model, to be deployed in future vehicles based on the Soar platform, will be trained with embedded language, enabling it to reason and interact via chat.
NVIDIA is working with Uber to roll out a Level 4 autonomous service within the next few years.