Building the Best Autonomous Brain

When I’m bumper-to-bumper in a sea of exhaust fumes and distracted drivers, it seems like autonomous driving can’t get here fast enough. Nor can the potential rewards that come along with fully autonomous vehicles, like far fewer accidents and mobility for people who struggle to get around on their own. To do my part, I’m focusing on how building the best autonomous brain for a car will get us there faster.

5 Things to Know About Autonomous Vehicles

Every day, we’re getting closer to the technology needed to power self-driving cars. But in-vehicle compute needs are complex, and autonomous driving algorithms are changing rapidly. So, the question is: What is the best long-term path to fast, safe decision-making? It all begins with the right compute for the right task. Here are five things you should know about the complex compute for autonomous driving.

 

It Takes More Than Deep Learning

Artificial intelligence is just one part of the story. And beyond that, AI is more than just deep learning. Yes, deep learning is key in teaching a car how to drive, especially when it comes to computer vision. But there will be several other types of AI at work in the fully autonomous vehicle, from traditional machine learning to memory- and logic-based AI. The fully autonomous vehicle will need a wide range of computing to support three intertwined stages of self-driving: sense, fuse and decide. Each stage requires different types of compute. In the first stage, the vehicle collects data from dozens of sensors to visualize its surroundings. During the second stage, data is correlated and fused to create a model of the environment. Finally, the vehicle must decide how to proceed. System designers need a flexible architecture to support all three stages, with an optimized combination of power efficiency and performance.

With a flexible, scalable architecture of CPUs, Intel Arria 10 FPGAs and other accelerators, our Intel GO automotive solutions portfolio leads the industry with a diverse range of computing elements that support all three stages of driving. But autonomous driving is much more than just in-vehicle compute; that’s why we offer a full car-to-cloud solution including 5G connectivity, data center technologies and software development tools to accelerate autonomous driving.
Smart AI consists of sensing, fusing and deciding.

 

No Fixed Architecture Can Keep Pace

Before system designers can achieve level four and five driving automation, they must determine how to best use different compute elements within the system to support each type of workload.

No fixed architecture can keep pace with the speed of innovation in AI and system design. Automakers and suppliers will need to be ready to change system designs down the road. Whether it’s to incorporate new algorithms or completely rethink compute to accommodate new workloads, system designers will need a flexible, scalable architecture. Simply put, they need interoperable and even programmable compute elements that don’t require them to start from the ground up every time they want to incorporate a new feature. With a flexible architecture of CPUs, FPGAs and other accelerators, future-ready solutions offer a diverse range of computing elements that can accommodate designs that may change long after hardware and vehicle design decisions have been made.

 

Driving the Future

Now is a time of tremendous opportunity as we continue to research and respond to the transformational changes before us. From powering Stanford University’s robotic car to serving as a premier board member of the University of Michigan Mobility Transformation Center’s Mcity, Intel is working alongside world-renowned research teams to understand the way people interact with connected cars. Intel has built autonomous vehicle labs in Arizona, California, Germany and Oregon. Here, we’re working hand in hand with our ecosystem partners to optimize customized solutions, road-test autonomous vehicles, and work toward common platforms that will speed broad industry innovation for the promising road ahead.

Learn more about the road to autonomous driving at intel.com/automotive. To stay informed about Intel IoT developments, subscribe to our RSS feed for email notifications of blog updates, or visit intel.com/IoTLinkedInFacebook and Twitter.

Jack Weast

About Jack Weast

Jack Weast, Principal Engineer & Chief Architect of Autonomous Driving Solutions, Intel / USA: Jack Weast is an industry recognized innovator and change agent in the adoption of modern Information Technologies in non-IT industries. Jack is the co-author of UPnP: Design By Example, is the holder of numerous patents with dozens pending, and is an Associate Professor of Computer Science at Portland State University.

Comments are closed.