A new Artificial Intelligence (AI) computing platform promises to be the breakthrough that will make autonomous vehicles affordable and practical without prohibitive infrastructure investment by governments.

American technology company, Nvidia, has developed an open AI car computing platform that enables automakers and their Tier 1 suppliers to accelerate production of automated and autonomous vehicles. Called Drive PX 2, it scales from a palm-sized, energy-efficient module for AutoCruise capabilities to a powerful AI super-computer capable of autonomous driving.

With a chip that processes 24 trillion bits per second, this unit is set to replace all computers on board current vehicles, at a fraction of the cost. Combined with laser-guided radar mapping, the platform allows vehicles to ‘see’ far more than any human driver can.

The new single-processor configuration of Drive PX 2 for AutoCruise functions – which include highway automated driving and HD mapping – consumes just 10 watts of power. Also, it enables vehicles to use deep neural networks to process data from multiple cameras and sensors.

Drive PX 2 can understand in real-time what’s happening around the vehicle, precisely locate itself on an HD map and plan a safe path forward. It’s the world’s most advanced self-driving car platform – combining deep learning, sensor fusion and surround vision to change the driving experience.

The scalable architecture is available in a variety of configurations. These range from one passively cooled mobile processor operating at 10 watts, to a multi-chip configuration with two mobile processors and two discrete GPUs delivering 24 trillion deep learning operations per second. Multiple Drive PX 2 platforms can be used in parallel to enable fully autonomous driving.

With a unified architecture, deep neural networks can be trained on a system in the data centre and then deployed in the car.

Sensor Fusion

Drive PX 2 systems can fuse data from multiple cameras, as well as lidar, radar, and ultrasonic sensors. This allows algorithms to accurately understand the full 360-degree environment around the car to produce a robust representation, including static and dynamic objects. Use of Deep Neural Networks (DNN) for the detection and classification of objects dramatically increases the accuracy of the resulting fused sensor data.

Nvidia AI platforms are built around deep learning. With a unified architecture, deep neural networks can be trained on a system in the data centre and then deployed in the car. Nvidia DGX-1 can reduce neural network training in the data centre from months to just days. The resulting neural net model runs in real-time on Nvidia Drive PX 2.
Nvidia DriveWorks is a Software Development Kit (SDK) that contains reference applications, tools and library modules. It also includes a run-time pipeline framework that integrates every aspect of the driving pipeline, from detection to mapping and localisation to path planning to visualisation.

Building partnerships

Nvidia’s partners in the car business include Mercedes, Audi, Bosch and Tesla, while Nvidia CEO, Jen-Hsun Huang, has announced that Toyota will use Nvidia’s Drive PX super-computers for autonomous vehicles.

Those cars will debut in the market in the next few years, Mr Huang said. Nvidia made the announcement at its GPU Technology Conference in San Jose, California. Mr Huang is confident that everything from commercial airliners (Airbus is designing one) to delivery trucks will be automated with a variety of technologies, including full auto pilot, mapping to assist the driver in a co-pilot function, and something called guardian angel. With guardian angel, the car detects hazards around you and warns you of an emergency, like someone else running a red light as you are about to enter an intersection.

Xavier uses a custom ARM64 central processing unit and a 512 Core Volta graphics processing unit (GPU). The chip is designed to be programmable and low power, and it will be able to run the software for self-driving cars. Nvidia is open-sourcing the Xavier deep learning architecture software, beginning in September.