Self-driving cars need AI, and we delivered big on AI at the GPU Technology Conference in Silicon Valley this week.
During his keynote, NVIDIA founder and CEO Jensen Huang made a host of announcements that will change how the automotive industry designs, builds and drives its cars.
Chief among these was the unveiling of Volta, the world’s most advanced GPU architecture, which will fuel the forthcoming DRIVE PX Xavier AI car supercomputer. Xavier is a complete system-on-chip that integrates a next-generation CPU, Volta GPU and our new Tensor Cores.
Toyota Chooses NVIDIA DRIVE PX
Huang also announced that Toyota, legendary for its high standards and priority on safety, has selected NVIDIA DRIVE PX for autonomous vehicles.
Toyota will use NVIDIA AI hardware and software for autonomous driving systems that will enhance the capabilities of autonomous driving systems planned for market introduction within the next few years.
AI Cars in the House
From the convention center lobby to the exhibition hall, GTC featured a variety of AI-enabled cars. The NVIDIA booth included BB8, our self-driving test car that uses deep neural networks (DNNs) to learn how to drive.
In addition to PilotNet — a DNN that controls steering — BB8 uses LaneNet for detecting lane markings, DriveNet for detecting vehicles, pedestrians and signs, and OpenRoadNet to detect the drivable area in front of the car. These DNNs work together to let BB8 safely drive in all kinds of scenarios.
A huge Peterbilt truck dominated another section of our booth. PACCAR — which manufactures the Kenworth, Peterbilt and DAF lines of trucks — has developed a proof-of-concept self-driving truck with SAE Level 4 capability built on NVIDIA DRIVE PX 2 technology, trained on deep neural networks.
There are currently 300 million trucks worldwide, driving over 1.2 trillion miles annually. We’re working to make every single one of them safe with DRIVE PX.
In GTC’s lobby, we featured cars from our partners Udacity, AutonomouStuff and Audi. Originally demoed at CES 2017 driving itself around a changing course, the AUDI Q7 Concept integrates BB8 technology for end-to-end deep learning.
The AutonomouStuff car is essentially a DRIVE PX 2 on wheels. The car comes configured with our AI supercomputer, loaded with NVIDIA DriveWorks, and comes pre-wired for sensors.
In addition to showing its self-driving vehicle, Udacity presented a session about their online Self-Driving Car Engineer Nanodegree. The nanodegree covers self-driving car topics ranging from deep learning and computer vision to sensor fusion and localization.
Self-Driving Car Sessions
Attendees had a great chance to dig into the details of autonomous vehicles during GTC’s 50+ auto sessions. From the role of deep learning in motorsports, to the power of HD maps to enable autonomous driving, our speakers covered it all.
Raquel Urtasun, who leads Uber’s team developing AI for autonomous cars, shared a variety of findings from her group at the University of Toronto. Speakers from Argo AI/Ford and Mercedes Benz highlighted how major automakers are bringing self-driving cars to life. And a host of startups and mapping companies — including Affectiva, MonoDrive, robotTuner, Zenrin, HERE and TomTom — spoke about innovations in the autonomous driving ecosystem.
NVIDIA also hosted several sessions on our cloud-to-car platform for self-driving vehicles. We highlighted the capabilities of DRIVE PX 2, our open AI car computing platform, and demoed examples of the reference applications, tools and library modules in the DriveWorks SDK.
For a hands-on experience, we hosted an instructor-led lab where attendees learned how to integrate various sensors, and then use DriveWorks modules in their own custom code or applications. This lab highlighted the open nature of DriveWorks and demonstrated specific techniques for enhancing its computationally intensive algorithms for object detection, map localization and path planning.
Transforming the Automotive Ecosystem
During the keynote, Huang also announced technologies that could revolutionize how the automotive industry designs and manufactures cars.
The demo took the GTC audience inside the design review of the all-carbon-fibre $1.9 million Koenigsegg Regera supercar. They watched engineers explore the car at scale and in full visual fidelity in the Holodeck, and consult on design changes in real time.
Huang also announced the new Isaac robot-training simulator, which will allow robots to be trained in the virtual world before they’re deployed in the real world. Working at super-human speeds, Isaac will enable teams to simulate the driving environment for the training and testing of autonomous vehicles. As well as the factory robots that help build cars all over the world.
Together with our partners, we’re changing the way people drive, and empowering vehicles to drive themselves. Join us at GTC Europe in October to see what’s coming next.