The Rush to Autonomous Automobiles: Are self-driving vehicles really just around the corner?

in Community

By Brian Douglas

In the Bay Area and especially in Silicon Valley, we’re pretty much ground zero in the frantic race to develop autonomous vehicles. Most of us have spotted self-driving test vehicles, some of us have been driven around in one and I’ll bet more than a few Climate readers are working on self-driving technology. Nearly every major automaker and Tier One supplier have research facilities in our tech-rich area. VW Group, the automotive giant that produces Audi, Bentley, Bugatti, Lamborghini, Porsche and Volkswagen has its Silicon Valley technology offices around the corner from Oracle’s headquarters.

A quick survey of friends and family about self-driving cars might uncover what I’ve discovered. Most people are perfectly happy to keep control of their vehicle in lieu of turning that task over to microprocessors and software. Perhaps that’s because we’ve all experienced a computer crash and we would rather have that phenomenon occur on our desktop than at highway speed.

When Deloitte’s Global Automotive Group studied autonomous driving attitudes worldwide, it discovered that the majority of drivers in China and India are fond of full self-driving. But in most other developed nations, not so much. American, Japanese and German drivers were the least enthused about surrendering control to the machine. Interestingly, these are the same countries with automakers developing the majority of autonomous systems.

Of course, self-serving is a great motivator of self-driving, so Uber and Lyft have both jumped into autonomous development to eliminate the cost of drivers. And both rideshare companies engaged Silicon Valley technology firms as well as legacy automakers to help with development.

Meanwhile, Tesla announced that its Autopilot, a level 2 autonomous system, could accomplish some self-driving tasks and suggested that its full self-driving system was well under development. So even though consumer demand has been less than lukewarm, suddenly billions of dollars have been poured into autonomous vehicle research in a world competition that is reminiscent of the mid-20th century space race.

On the surface, a self-driving system seems pretty straightforward. Many of today’s cars have “adaptive cruise control” systems that can maintain a set distance between vehicles, even stopping and starting in stop-and-go traffic. Some are equipped with “steer-by-wire” and lane-keeping systems as safety enhancements and a few have radar surround systems to report the vehicle’s environment with other traffic. If you simply integrate this hardware and software into a computer, shouldn’t that do it? Nope, not even close.

Self-driving vehicles gather information about their surroundings through a battery of cameras, radar, sonar, GPS and LIDAR (pulsed laser light) and process this data in an onboard supercomputer. In its self-driving system, NVIDIA’s DRIVE PX Pegasus crunches 320-trillion operations per second, a performance on par with a 100-server data center. Intel’s less ambitious system estimates 4 Terabytes of data generated in an hour and a half of driving. If you thought today’s vehicle electrical systems were complex, just wait a few years.

Not every bit and byte of self-driving has to be crunched in each car. Connectivity to other vehicles along with road and highway infrastructure would go a long way to enable self-driving without the heavy lifting of Big Data and Learning AI (artificial intelligence). But no one seems to be patiently waiting for governments to take a leadership role. In fact, some of us would be pleased just to see the potholes fixed.

Some reasonable people wonder why we can’t simply adapt aircraft autopilot technology to automobiles. After all, some big planes today can even land autonomously. That’s true, but the Auto-land systems in commercial and large business jets connect with runways capable of a Category III approach, they have far less traffic to avoid and always have a thoroughly trained professional ready to reassert control. What’s more, they’re backed up with two or three redundant systems and cost at least a quarter million dollars.

I had my first autonomous riding experience 14-years ago in Stanley, a VW Touareg SUV created with Stanford engineering students with help from VW labs and Intel that had won the Defense Advanced Research Projects Agency Grand Challenge. Stanley’s entire roof was populated with radar and LIDAR devices and the spacious interior, from the front seat backs to the tailgate, was filled with rack-mounted electronics. We were driven slowly around the campus by programmed microprocessors. Back then, that was a real hoot.

Fast forward to today, and I’ve enjoyed an autonomous ride in an Audi Q5 around the streets of Menlo Park in traffic, merged onto 101, made a safe exit and driven back to Aptiv’s Silicon Valley laboratory. Aptiv (formerly Delphi) is a major tier one supplier to the auto industry and it’s fully autonomous Audi looked completely stock. No “chicken bucket” or sensors on the roof and there was room for luggage behind the second-row seating. The new electronics, with supercomputer power, were tucked under the flat floor in what was once the spare tire compartment.

The route we took was pre-scripted by engineers, so while there can be surprises from non-autonomous fellow motorists, the ride is nearly always adventure-free. But for self-driving testing, there’s always an engineer with a California Autonomous driving license (I want one of those!) ready to take charge. That’s a good thing, because self-driving algorithms can only know and learn scenarios they’ve been programmed or experienced through machine learning. What’s more, fully autonomous vehicles are taught to obey speed limits to the number and be very cautious when merging. You can imagine what a hit that is in today’s Silicon Valley’s driving culture.

When I asked my fellow passenger, the engineer seated behind the wheel, what weird things he’d experienced in the program, he shared what could have been a tragic story. An initial test ride around Mountain View went just fine and the vehicle was making its exit. The computer saw a 25 mph sign midway down the exit ramp and slowed immediately to 25. That was quite a surprise to the commuter in his BMW following the autonomous Audi. Fortunately, the engineer’s quick reflexes and training saved the day. Back at the lab, the software was reprogrammed to slow gracefully at that exit.

I had to ask my engineer seatmate (I can’t really call him the driver) how often he was flipped off by fellow motorists. His answer was that the most frequent response to his vehicle’s cautious, law-abiding manners was simply the Silicon Valley stare. That’s when someone pulls beside and sees what looks like a normal Gen X driver and thinks “Really?”

With all the time and treasure pouring into autonomous driving in every developed country, we will certainly see these vehicles progress from expensive science projects to mass production. But like earthquakes and the Richter Scale, moving one number up on the autonomous scale is a big deal. As smart as computers seem, the human brain leaves them in the dust in cognitive ability behind the wheel. OK, assuming that person is reasonably skilled. So we’ll see autonomous vehicles begin to populate urban areas. And we may soon experience the cool self-parking described earlier. And hopefully some of us will still be able to own our vehicle and drive when we choose. Then there’s the legal and moral issues of who or what’s to blame when things go wrong. But that’s a different story …