By Kathy Winter
I get to visit many metropolises for my job – Munich, Detroit, San Francisco, Tel Aviv – and I absolutely love the hum of the big city. The activity is invigorating – there's so much to see and do. Yet that vibrancy loses its appeal when I'm behind the wheel of a car. In one city block, someone will honk at me, another car will cut me off, a siren will blare, a delivery truck will block my lane, a garbage can will fall off the curb, a pedestrian will dart out in front of me – not in a crosswalk, mind you – and someone on a bike will come at me from the wrong direction. That short drive requires every ounce of attention I can muster.
Short city blocks represent the longest leg ahead on my autonomous journey at Intel – not in distance, but in engineering complexity. Most of today's self-driving cars – or cars with Level 3 conditional automation – have already shown they can safely travel long distances on a highway. Consider the 3,400-mile cross-country trip my team and I engineered two years ago at Delphi. Ninety-nine percent of the time we were in self-driving mode and safely navigated all kinds of weather – even staying on track when the lane markings disappeared. But things get much trickier when you leave the well-marked highway.
The reason for this is simple: By their very nature, city streets are exponentially more complex than highways. The variety of objects encountered on a highway drive is relatively limited: cars, motorcycles, trucks, street signs, trees and bushes, guard rails, and a few other possible options. Leave the highway and much more is added to the mix. Humans for example – an infinite variety of humans walking, running, riding bikes, riding skateboards and scooters, riding hoverboards, going the wrong way, jumping out of cars, jumping into taxis.
Before cars can be truly driverless – no steering wheels, no pedals, and no humans operating as a failover mechanism – we have to engineer car brains that can handle the complexity of a dense urban environment. As defined by the Society of Aerospace and Automotive Engineers (SAE), this means Level 5 full automation. And that is by no means an easy feat.
While the incremental naming of the different levels of automation suggests an incremental increase in capability, the deltas are more in terms of magnitude. Level 5 driverless cars will need exponentially greater sensing and decision-making capabilities than Level 3 cars. In other words: exponentially more computing intelligence and performance – monumental jumps in the complexity of the algorithms and associated computational throughput with almost zero delay, which means significantly lower latency. While technology exists today that can meet all of these needs, no company has yet been able to scale a computer into a thermal or power envelope that meets the needs of the industry.
Despite these challenges, Intel is alone in being able to reach this computational value and performance scale across the car, the network and the data center. Intel recognized this multifaceted challenge needs a multifaceted toolkit – with an optimized mix of scalable CPUs that can extend to data center server-class products, FPGAs and, yes, artificial intelligence. And Intel is the only tech company that offers a complete, end-to-end toolkit that spans the full technology challenge car-to-cloud to make fully autonomous driving possible.
Most importantly, driverless technology promises us the potential to save hundreds of thousands of lives and grant mobility to all – not to mention saving us the stress of driving across a vibrant and busy city or down a long and boring highway. It will take the entire depth and breadth of Intel's portfolio plus its key partners to travel the distance between the Level 3 cars we are seeing today and the Level 5 capabilities that will finally remove humans from the driver's seat tomorrow. This is one trip where I'm excited to be in a passenger seat, with no one behind the wheel.
Kathy Winter is vice president and general manager of the Automated Driving Division at Intel Corporation.
This is the sixth in an occasional series of Intel newsroom editorials related to autonomous driving. To comment or reach Kathy directly, email firstname.lastname@example.org.