How NVIDIA Is Strategizing to Build an Autonomous Vehicle

NVIDIA’s end-to-end autonomous car solution

NVIDIA (NVDA) is looking to make the concept of a fully autonomous car reality by developing an end-to-end platform, from architecture to HD (high-definition) maps. It aims to develop a complete software stack and update it on a continuous basis. On top of that, it hopes to make this stack open, which means anyone can contribute to the stack. At NVIDIA’s 2017 Investor Day, Automotive senior vice president Rob Csongor explained his strategy to achieve Level 5 automation.

How NVIDIA Is Strategizing to Build an Autonomous Vehicle

What is NVIDIA’s strategy to develop an AI car?

NVIDIA’s Drive PX 2 AI platform collects data from sensors, cameras, lidar, and radar, develops a 3D map of a car’s surroundings, localizes itself on an HD (high-definition) map, and forecasts potential risks while driving. The chip supplier is working with 35 vehicle makers, 24 Tier 1 suppliers, 37 start-ups, and ten HD mapping companies to test prototype vehicles. Tesla (TSLA) will launch NVIDIA Drive PX 2–powered cars by the end of calendar 2017.

Developing deep neural networks

NVIDIA’s first step towards developing an AI car is to study cars equipped with its Drive PX platform by collecting data generated by them and processing it. The company uses this data to train DNN (deep neural network) systems.

Role of HD mapping companies

Creating an HD map is a complex process, as it requires a huge amount of point cloud processing and AI. NVIDIA has partnered with several mapping companies to create HD maps for self-driving cars.

These HD maps and training results have to be sent to the car on a real-time basis as OTA (over-the-air) updates. This information is processed by the vehicle, which would be equipped with a supercomputer and full software stack that would handle detection, localization, path planning, driving, and AI.

Xavier platform

In 2016, NVIDIA launched its Drive PX 2 platform, which featured four processors. The company will replace this platform with its more powerful, fully programmable, AI-based Xavier platform, which is specifically designed for Level 4 autonomy.

During the question and answer round on NVIDIA’s Investor Day, CEO Jensen Huang stated that NVIDIA’s Xavier platform has structural advantages over Intel’s (INTC) platform, which is a mix and match of CPUs (central processing units), FPGAs (field-programmable gate arrays), and Mobileye’s (MBLY) custom ASICs (application-specific integrated circuits). Next, we’ll see how NVIDIA aims to bring AI to robotics.