Around mid-October the second DARPA Grand Challenge race had gear heads and bit heads glued to their Web browsers watching robotic cars traverse the Arizona desert. Unlike last year, many of the self-guided trucks actually made it to the finish line. It was a milestone event for robotics, for automotive electronics, and for private ingenuity. Among other things, the government-sponsored event showed that a specific and well-defined challenge can bring out the best minds in private enterprise and universities.
Although the robot race was conducted for military purposes—the Pentagon wants to field real automated vehicles soon–I have no doubt much of this technology will find its way into civilian cars before long. There's much here that can be applied to lane-departure warning systems, collision-avoidance systems, navigation systems, and all sorts of other “peace dividend” applications.
That's not unusual. Military technology often filters down to everyday use. The GPS satellite navigation system, for example, was designed for intercontinental weaponry but is now ironically in the hands of environmentally conscious anti-establishment hikers and backpackers throughout the country. The oft-cited example of Teflon, however, was never designed for the space program. DuPont was selling Teflon before NASA was ever formed.
How will the Grand Challenge droids' technology make its way into more pedestrian vehicles? I suspect the image-recognition and obstacle-avoidance algorithms will prove the most valuable. Image recognition requires a lot of computing horsepower, so to speak, along with some serious applied mathematics. Given enough production volume, it starts to make sense to design custom chips to do this work, rather than throw code at generic microprocessors. Those SUVs lumbering through the wilderness were the prototypes, but now that those systems have been proven to work engineers can start cost-reducing them. Custom image-recognition processors that embody much of the entrants' intellectual property can't be far behind.
The same goes for obstacle-avoidance technology, which will play a role in future active-safety systems. Today DaimlerChrysler has a comparatively crude adaptive cruise control that maintains station-keeping distance from the car ahead and can apply the brakes when the closing speed increases suddenly. But true collision-avoidance will be much more tricky and sophisticated. Would the vehicle steer clear of an obstacle? Or would it simply apply the brakes with enough pressure to avoid hitting something directly ahead? Would it pay attention to other vehicles closing from behind and adapt to avoid rear-end or three-way collisions? There's no single answer to these questions because each situation is a bit different.
Legislation and consumer acceptance will likely determine the evolution of these systems. Now that we know it can be done (albeit in a large and ungainly way) we need to decide if it should be done. Personally, I like driving my car, not riding it. The best safety measure may be better driver training combined with stricter licensing. Instead of idiot-proofing the car let's just improve the idiots. Now would be a grand challenge.
Jim Turley is editor in chief of Embedded Systems Design magazine.
Nice upbeat article. However, it is unlikely that we will see any of this self-driving technology in our cars anytime soon. Not because it's unimpressive or useless but because its complexity is its Achilles' heel. Software unreliability is proportional to complexity. Unless safety-critical software can be guaranteed safe and reliable, it cannot be released to the public on a massive scale.
As an example of what could go wrong, this automated bus in the Netherlands recently collided head-on with another on a one-way lane: www.2getthere.nl/2GT3.php
Here are some pictures of the crash: www.geenstijl.nl/paginas/connexxioncrash.html
Unless we can find a way to guarantee that complex programs and systems are defect-free, I am afraid that computing will never reach its true potential. Fortunately there is a solution. We must abandon the algorithmic model of software construction and adopt a non-algorithmic, signal-based, synchronous model. Additional details can be found here: www.rebelscience.org/Cosas/Reliability.htm
– Louis Savain
This article reminded me of when, in 1993, the company I worked for in Scotland, called Ferranti, designed and built an in-car map display and navigation system with GPS.
It worked, but not in the confines of a city, filled the boot/trunk full of electronics, and cost five times the price of my car. Not surprisingly, it never featured in our product portfolio.
It is interesting to see the commercial equivalents on the market today, but I am sure there are some lessons here that are best learned and not repeated.
– Martin Allen
We've done just fine without the “Grand Challenge Technology” embedded into our automobiles. There are no statistics indicating that we need this technology in our vehicles. But I'm sure marketing and the big auto makers will convince the public (and lobbyists) that we need it! I'm sure “Soccer Mom's” will buy into it! Just another mark up on auto sales as I see it. Just like anti-bacterial soap, we've done just fine without it. Now it's marketed like crazy.
Maybe this technology can be embedded into Border Patrol Uniforms, to prevent head on collisions with all of the people crossing into the US!
– Steve King
Kaman Aerospace, Electro-Optics Development Center
Steve King wrote: “There are no statistics indicating that we need this technology in our vehicles.”
You are mistaken, in my opinion. We need this technology desperately. If we could eliminate the need for human drivers, we would save millions of lives around the globe. It would also open up safe tranportation to millions of people who cannot drive due to age or some physical handicap.
In addition, we would save a huge amount of energy. The reason is that, with driver-less vehicles, cities could ban private automobiles altogether. City dwellers would be given a wireless pager with which to summons a vehicle at the click of a button. The nearest automated vehicle would then drive itself to the customer's location and safely take them to their destination. There is no need to have so many vehicles on the road since most of them are idle most of the time.
– Louis Savain