We have a long way to go before we see autonomous vehicles on the roads in any number.
Is the autonomous vehicle (AV) going to be widely distributed in our lifetimes? Sure, just as soon as you eliminate human involvement, complacency and ignorance (was that repetitive?) from development. Until then, AVs are just going to be a rarity on the roads.
The electronics world may differ. There are a lot of companies gearing up to sell a lot of components to go into these cars, but the automotive industry, with the exception of Tesla, is starting to hedge its bets. Both GM and Ford, who promised to roll out their autonomous products this year have delayed the rollout indefinitely. Ten years is now the optimistic prediction. Twenty is more realistic.
It’s not the hardware portion that’s the problem, though. Well, not completely. Arm’s autonomous vehicle coalition was founded, largely, to solve the compute problem posed by artificial intelligence, most versions of which are housed in data centers that would overwhelm the myriad of computers in contemporary vehicles. So they are working on it.
The problem is they have no idea what exactly is going to end up in their compute systems in each car, because the people building the artificial intelligence (AI) have come nowhere close to perfecting it.
Take Ford’s announcement to delay the roll out. One of their test cars was on a busy Manhattan street when it encountered a cyclist coming between the lanes opposite the flow of traffic. The AV did not know how to respond to this inference and swerved into a car in another lane. That’s when Ford realized that the level of human variables a vehicle’s AI would have to deal with had not yet been found.
That wasn’t the only case. As EETimes Junko Yoshida pointed out a few weeks ago, the Tesla “summon” feature has trouble dealing with parking lots and shadows. CBS Radio tech correspondent Larry Magid recently posted a video on Facebook to demonstrate the feature and when the car came to a shadow cast by a tree, it came to a dead halt. It couldn’t handle the variation in lighting.
It is not, however, a question of whether vehicle autonomy will ever be viable. As Rik Turner, Ovum’s principal analyst put it, “We’ve had certain levels of vehicle autonomy since the 1980s when cruise control was first introduced. We will continue to creep ever closer to the ideal situation but the human stupidity factor can never be underestimated.”
Turner sees the application of level 4-5 autonomy best applied to public transit rather than private vehicles and, in fact, the longest and most successful application is Dockland Light Railway system in east London that has been fully automated since the Margaret Thatcher administration. Even here this is where the human factor delays adoption.
The Bay Area Rapid Transit (BART) in the San Francisco Bay Area was originally supposed to be an autonomous system when it was founded in 1972, but strong labor unions at the time put then kibosh on the idea. That could change soon. After several expensive strikes and the lack of funding to do proper maintenance on the system in the past decade, pressure is mounting to activate autonomous technology that already exists in the system framework and divert salaries and benefits to the engineering teams.
Turner pointed to an extensive special report from The Economist (Sorry, but it’s behind a paywall) that surmised that, There is a better chance for AV success in the United States then Europe specifically because of the lack of effective public transport in America, but in the realm of collective, elective transport than in personal vehicles. This has some significant merit. AVs programmed to take a specific route on a regular basis are less likely to encounter aberrant human activity in unfamiliar areas. A six-person AV could pickup a car pool of individuals traveling from one restricted geographical area to another with less problem than a private vehicle traveling different routes on a daily basis.
Josel Lorenzo, vice president of product development at smart security startup Axiado Corporation, concurred. Lorenzo has been an AV product team member at several companies including Intel. He sees a basic disconnect between how humans interact naturally and how an AI would view a similar situation.
“Let’s say two human drivers reach a four-way intersection at the same time. There are multiple variable, outside of the vehicular laws, to determine who will have the right of way. It could be a nod of the head, or waving someone through or any other non-verbal cues.”Lorenzo pointed out that an AV will only do what the law says should happen and there is nothing stopping the human driver from doing what the AV does not expect. Nothing is normal on the roads when humans are involved.
“In certain areas, you may have to have a law that says you cannot enter the area unless you are in autonomous mode so all vehicles are autonomous.” Lorenzo suggested. “It is just safer to not mix autonomous and human-driven vehicles. In fact, we may not want to have individual ownership of cars in the near future.”
And that’s only if we can resolve the issue of securing the data used in the vehicles.
Matthew Rosenquist recently left his post as senior cybersecurity analyst at Intel and is now a “gun for hire” consultant to several organizations and has been following the AV development with a kind of dark glee.
“This is a really juicy subject that needs to be unpeeled like an onion,” he said. “the basic problem in AV security is we don’t know what we don’t know. The automotive and electronics industries have underestimated the challenge of security in AVs and that lack of understanding didn’t stop them from making commitments. Now they have set expectations in the public and media that they can’t meet. The level of complexity is just now being understood.”
Rosenquist said that as cars were just starting to be connected to the internet through a variety of electronic control units, the general consensus was that security is very important. “However, when security concerns start impacting development schedules in any technology development, security tends to be de-prioritized.”
The basic problem with AV security is the technologists leading development are trying to develop a framework that looks like a personal computer. They are minimizing the fact that those computers are hacked constantly believing that they can fix the problem with monthly updates.
“The stakes are different,” he stated. “If you get the blue screen of death at 70mph on a freeway you can’t reboot your family. And with a completely autonomous car connected to your home network, a hacker can open your garage door, start your car and drive away with it while you sleep.”
The audio interviews that informed much of this article are linked above at my Crucial Tech podcast. Also, follow Junko Yoshida’s coverage of artificial intelligence and autonomous vehicles over at EETimes. I would especially appreciate dissent in the comments below.