The infamous Toyota case may be behind us, but cars still have bugs, and there are many more to come.
We hear and see all sorts of predictions about the future of fully autonomous (self-driving) cars, ranging from "just a few years away" to "much further away than you think." These forecasts come from all sources: by people closely involved with projects; by techno-addicts who are enthralled by the promise of the "new" (often with good reason, but equally often not); by market researchers who sell high-priced optimistic reports; by vendors who hope to sell the many components needed; and by desktop pundits who really don't know much but it all sounds so good.
I'm hopeful but skeptical for a variety of reasons: cost (electronics don't come cheap), capacity (a trunk full of electronics doesn't leave much room for groceries), and perhaps most concerning, my experience and observation that completing the last 10% to 20% of a project takes 80% to 90% of the project's time, especially when it involves lots of software, many functions, and complex interaction among the many building blocks. I suspect that we'll be at Level 2 and Level 3 of the five-level autonomous-car hierarchy for a long time before we get to Level 4 and Level 5 (see "Autonomous driving levels 0 to 5: Understanding the differences"). We'll also need a good definition of what level of performance is "good enough." Will it be 95%, 99%, or 99.9999%? (We'll ignore the legal and liability issues in this assessment.)
Software bugs in vehicles won't end anytime soon.
It's a function of "where" as well: Putting a successful higher-level vehicle on the highway is a lot different than doing so in wild and crazy urban traffic. How will Level 5 cars handle poorly marked parking garages with no GPS signal available, the ones where even a human driver can't figure out how to get to the exit? I can't imagine a computer parking in Boston's Landmark Center garage with all of its poles.
There's another reason that I am skeptical. Today's cars are already heavily loaded with functions, features, and associated software for both internal functions as well as the driver console and display, and the results are uneven, to be polite. Even high-end cars have problems with frozen displays, unstoppable driver alerts, beeping, confusing controls, and more. I don't think that the average driver wants to take the half-day "functions and features" classes that some new-car dealers offer; also, even if you do become proficient on your new car, what happens when you travel and rent another make of vehicle?
You can argue that the problems drivers are seeing now will go away after cars become so smart that all they have to do is get in and say or type in where they want to go. I'd like to think that, but there will still be user displays, entertainment systems, alerts, and updates. In some way, the Tesla mode of downloading updates is both a blessing and a curse because it supports continuous improvement, but it allows constant updates with all the unexpected and often undesirable surprises that they bring.
Updates also allow engineers to subconsciously think that they can have users become the ultimate beta-testers, although that's not a viable approach for cars, obviously. I know that developers of autonomous vehicles, such as Google, have logged millions of miles in their test vehicles, and that's good. Still, testing a relatively small number of such cars, even for millions of miles, is not the same as testing millions of cars for even just a few thousand miles each, as problems that are on the edge of the bell curve may not show up in the former case. Several years ago, when the Microsoft Windows OS was "buggy" and the blue screen of death was a regular occurrence, the not-so-funny joke was that if you had a car designed by their software team, it would stop by itself every few miles and you'd have to restart it, bringing a whole new meaning to the word "crash." I surely hope that autonomous vehicles laden with "features" don't have that type of reputation problem for their first years.
How will this all shake out? I'll be honest, I don't know. I do know that if you look at the track record of most such predictions, we can easily conclude that the future arrives both much earlier and much later than was predicted and in a very different form than anticipated.
check back in five, 10, and even 20 years and we'll see how all of these predictions turned out. In the interim, I'll be happy with a car that is very good at doing its basic functions and not let peripheral functions get in the way.
— Bill Schweber is an engineer and technical writer who writes blogs for EE Times and EDN.com.