NASCAR For Nerds?
Looking at the crowds at the 2007 Urban Challenge, you'd be forgiven for wondering if DARPA has touched off another of the famous side-effects from its research projects.
The event was open and free to the public. While the teams and DARPA staff were present for the duration, the spectators came and went through the day, making it hard to judge the crowd. But I'd say at least a third and maybe half of the attendees were fans. Some seemed to have found a new kind of southern California entertainment.
In reality, the goals of DARPA and the tastes of race fans are in conflict. Safety and reliability aren't usually compatible with speed and risk. (There could be a future for autonomous bot races on the tube, however. Rumor had it that the presence of Discovery Channel talent indicated a forthcoming special or mini-series on the Urban Challenge.)
The Grand Challenges are like NASCAR in some important respects, however.
Not For Amateurs Anymore
Back in the day, 'stock cars' could be built by shade tree mechanics (and may have spent non-race nights running moonshine). Those days are long gone, and NASCAR is all business with lots of sponsorship money at stake now. That's happening here as well.
Of the teams that reached the finals of the Urban Challenge, only one - UCF - did so without major corporate sponsorship. The other shoestring efforts that were admitted to the competition fell out quickly during the qualifications.
It's a pattern common to technology. When a new area of research or products is broken open, the time and expense required to reach the 'edge' are relatively low. Those who date to the beginning of the microcomputer revolution will remember when anyone with a source of 8008 or 6500 chips and a modicum of funds could knock out a PC wannabe. A software 'product' took a few man-months. What we were actually producing were salable prototypes.
The Urban Challenge vehicles are also prototypes, and most teams will readily admit it. Sensors protrude, need to be cleaned periodically, and flake out from RFI, sun glare and dust. Server room rack mounts or Apple's consumer machines crammed into cargo areas are hardly milspec. The uniformed armed forces attendees on Friday were invariably polite, but I'm sure a number of them later had a good laugh considering how these machines would bear up in the heat, vibration and dirt of the sand box.
Miniaturization, environmental hardening, integration, testing and on and on. All of these are capital intensive, are required before seeing a return on the investment, and are beyond the charter or means of academic researchers and do-it-yourself teams. The barrier to entry is going up rapidly.
The Tale of the Logos
As you'll note, they are the very same logos. It was the automotive industry, both consumer and industrial, that was all over the place. Caterpillar sponsored three teams and there were CAT hats, T-shirts and logos wherever you looked. I'm sure the General Motors PR folks are exchanging high fives today - the GM logo covered about half the side of the winning Boss robot.
Your usual defense industry vendors were in evidence. LockMart, Honeywell, Teledyne - to name a few - have made contributions of subsystems and presumably people and cash. But their logos were small and their presence quiet.
Admittedly the automotives value brand equity, while defense is the ultimate inside selling business. But the real tale is in the composition of the teams. Here's an algorithm I tested at the event: Find someone wearing a T-shirt from a team that finished the Challenge, who appears to be over 30, and chat them up. If they aren't a university professor or staffer, then you've likely just located someone seconded from Volkswagen, GM, Cat or the like.
The clear lesson that comes out talking to these folks is that they see their Urban Challenge spinoffs reaching market well before there are convoys of autonomous logistics vehicles on the battlefield. Fully robotic cars are still a ways off, but semi-autonomous safety and convenience features - the next generation of self-parking - are already on the roadmap. Having participated in a few 70 to 80 mph 'freight trains' on I-5 while driving home, I'm all for that. One can easily foresee a not-so-distant future in which teenage and elderly drivers are required to engage autonomous safety features when at the wheel.
Fully autonomous systems will have an earlier home in industrial applications. Mining, maintenance and other 'dull, dirty and dangerous' (and often remote) jobs are square in Caterpillar's sights, and are a major part of the rhetoric coming from the winning Team Tartan. (Disclosure: I have an investment in this sector.)
The Urban Challenge may herald a transformation in both consumer and industrial automotive that's as profound as any defense meaning of the word.
A Convergence Of Platforms
For those of us in Silicon Valley and in hi-tech generally, a 'platform' is something largely built out of software. Think Linux, Google Web services, Windows or OS X. Microsoft's recent habits aside, we're accustomed to having the platform's functionality updated and augmented once or twice a year at least. Platforms have more or less well described functional interfaces that allow third parties to access their capabilities.
High tech investors pay a lot of attention to the creation and augmentation of these platforms. Their appearance and success is a sign that a market is reaching consensus on a set of sufficient functions to address its needs. Because it allows outsiders to tap and augment this functionality, the advent of a software platform means that the capital costs and other entry barriers of building a salable end-user product are ready to fall again. That can spell opportunity for startups. Innovations they spawn are often acquired, integrated and made generally available to platform users.
Having grown up in a GM family, I'm well aware that 'platform' has meant something very different in the automotive industry. It's a collection of structural, motive and control systems that will be the underlying basis for a number of superficially different vehicles. Rather than cycling in months, an automotive platform might turn over once every five years. Any 'interfaces' have historically been mechanical, always highly idiosyncratic, and generally unavailable to those outside the automotive OEM's supply and maintenance chains. The gulf between OEM and 'after market' components markets has been huge. Lacking the ability to wait out the lengthy platform cycle, and considering the risks of selling to a highly concentrated OEM market, private investors and consequent innovation have generally avoided this area.
The Urban Challenge represents one of two revolutions that will force the automotive market closer to the Silicon Valley notion of 'platform'. The other driver is the change in motive systems, and it's worth a quick detour. Hybrids are the first step toward what's likely to become all-electric motive power in many vehicles. Once the step to electric motors with 100% capability is taken, the source of the electricity is literally decoupled - the old fashioned drive shaft vanishes. Batteries, fuel cells, or hydrocarbon driven generation can be mixed and matched. The phrase for this in software architecture is "separation of concerns", and it's important not only for modularity, but to allow the modularized functions to innovate separately and usually more rapidly.
The Urban Challenge fits squarely into the second transforming theme, drive by wire. Accelerator linkages, shifters, brake master cylinders and even steering columns are slowly disappearing, replaced by electronic interfaces to distributed actuators. Which are in turn the perfect plug-in spots for the capabilities pioneered in the DARPA events. Add a similar data stream for sensor data, and the road for continuous improvement of safety and other autonomous functions is open.
We're probably never going to want to download new functions from the Internet into the family car. But if the military and government in general exercise some wisdom in mandating open interfaces when they begin to procure autonomous systems, we can avoid the era of 'fork lift upgrades' that occurred in avionics and allow flexible upgrades on fielded equipment. A separation of concerns of sensor array, actuators, and supervisory software will allow more independent innovation in both the civilian and military markets.
What's The Next Challenge?
DARPA's boss Tony Tether unsurprisingly dodged that question at the press conference following the Urban Challenge. If nothing else, his team now has weeks and months of pouring over captured data to find out what really happened from a systems point of view. How much of the outcome was robustness, and how much good luck? Until that's known, the real state of the research front is not. One can expect that the Boss, Junior and Odin may meet again, somewhere out of the public eye.
A fast redirect of the question to the heads of the three leading teams produced interesting answers that may shed some light, however. Red Whittaker of Tartan Racing is ready for an 'enduro', a long range road race forcing the bots to cope with darkness, rain, snow and dust - a test of the robustness needed for real deployment. Sebastian Thrun of Stanford's more car oriented team would like to push speed and cornering, moving beyond the limits imposed by the human senses and muscles. He mentioned an idea to mix bots and humans at a future Long Beach Grand Prix event. The Virginia Tech team would like to see collaboration between multiple vehicles, in convoying or other 'flocking' maneuvers. To the extent that the winner's ideas reflect the cutting edge of engineering, they may give some clue of what to expect if there's a fourth robotic DARPA Challenge in store.
DARPA's other major goal for the competition was to spur interest in robotics generally, particularly in hopes of attracting talent into a field that will become strategically important in a sense beyond the direct defense implications. Based on comments from Stanford and other schools, they are seeing increased interest, and the Challenges have stimulated growth of a larger robotics community.
Some of the most fatuous commentary and questions from the mainstream media portrayed the Challenge bots as potential HAL 9000s or general purpose artificial intelligences. They are nothing of the kind. They have more to do with prosaic signal processing techniques like Kalman filtering, enabled by lots of processor power, than they do with any sort of general 'reasoning' power. As an articulate blogger from the Cornell team put it:
"I guess the thing that's difficult for passers-by to realize is that these cars lack the ability to learn: they only know exactly what they've been programmed. If one of these cars changes lanes or makes a turn smoothly, it's because some dedicated zealot stayed up late one night tuning the crazy thing until it looked just so."
It's an easy trap to fall into, though. Even Dr. Tether admitted that he ended up attributing personality to the bots, as did their attendant human chase drivers, and my own post on their body language. Personality and intelligence are the only mental models and language that humans have for that degree of autonomous behavior.
That straightjacket of language hides a major change in the direction of robotics development. Historically, the image of the robot in both fiction and research has been what the classic movie Blade Runner called replicants. Substitute humans, in other words. That makes for better stories, and an endlessly challenging - perhaps impossible - research agenda.
Boss, Junior, Odin and the rest, along with Roomba's, Japanese eldercare bots and many others are something else. They are symbionts, creations we have made to live alongside us and extend our capabilities. They do not substitute for the human. It remains our judgement on what medical treatment is required, how often to mow the lawn, or what the rules of engagement require on the battlefield.
The inescapable physical presence of a bot like the TerraMax hides something else: Many of the symbiots that are beginning to surround us are out of sight, either virtual or just thought of as useful features of a product or service. TiVos, Amazon recommendations, a car's skid control system, all act on our explicit or inferred command. We're all becoming part of robotic systems, just as much as any future commander of a convoy of DARPA-derived logistics bots.