Monday, November 29, 2004

The Technology Squeeze: Truck Drivers

DARPA’s Grand Challenge gets more press than it deserves. It’s an entertaining competition, and there really is some great engineering work being done, but like most of DARPA’s projects, the Grand Challenge is about planting thousands of seeds in the hopes that something will turn out to be a useful technology in the long term. A short-term incubator it is not, but the competition was designed to meet Congress’s goal of making 1/3 of its operational combat vehicles unmanned by 2015. Now, 2015 may not sound short-term to some readers, but given the historically slow procurement cycles in the U.S. military, I think this is going to be a very difficult target to reach. 1/3 is a LOT of vehicles.

I think the Grand Challenge is an unrealistic approach to meeting that goal because it is spurring development in systems that allow for completely autonomous driving. The level of artificial intelligence required for this is huge, because it is so difficult for a system of cameras, radars, sonars, laser rangefinders, and their digital processors to distinguish between a bush, a boulder, and a brown paper bag. At least one team in the previous round of competition resorted to instructing the vehicle to not to drive over any bushes at all. Imagine a combat vehicle unwilling to risk driving over a shrubbery!

Not only is complete autonomy for ground vehicles unrealistic for large-scale deployment in the near term, but pushing for unmanned combat vehicles may not be the best way of meeting Congress’s ultimate purpose behind the goal, which was presumably to save lives. Recent counterinsurgency experience in Iraq–the kind of warfare our military expects to be the norm in coming decades—suggests that non-combat vehicles are the more dangerous place to be. Every guerilla knows that the supply chain is the soft underbelly of any large military. Why go after M1 tanks when unarmed, unarmored, and often unescorted semi-trucks provide ample opportunity to increase the body count on the evening news and reduce the ability of the U.S. Army to extend its reach into insurgent strongholds? Now that the Army is scrambling to better armor and escort its truckers, perhaps it’s time to consider alternatives to having a human driver in every cab; alternatives workable in the short term. Semi-autonomous alternatives.

Way back in 1995, a robotic driving system called RALPH drove a vehicle from Washington D.C. to San Diego, CA in control 98.2 percent of the time at an average speed of over 62 mph. Efficiency dictates that just about anywhere robots could work successfully 98% of the time, they should be allowed to do 98% of the work. As I've pointed out in my previous posts on robotics in the workplace, this is made possible by keeping a human in the loop. In RALPH’s case, the human in the loop was the system's designer, a doctoral candidate in the driver’s seat by the name of Todd Jochem (he subsequently earned his Ph.D.)

Obviously it wouldn’t do much good to give Army truckers autopilots that needed the human in the loop to be sitting just inches away. But what if truck convoys were actually chains of semi-random X’s and O’s, like XOOXOOOXOXOOXXOOO, where every ‘X’ is a human-driven vehicle and every ‘O’ is a robot-driven vehicle that mostly just shadows the vehicle in front of it as though following a short trail of breadcrumbs? This would get around the bush/boulder/bag problem because the robot trucks would only drive over the tracks of the human trucks. It would retain the intelligence of human drivers while easily eliminating a third of them. Just tint the windows of all the trucks, and stick dummies in all of the unmanned ones so that the guerillas can’t easily tell which trucks have fleshy wetware on board.

But what about that lead vehicle? It might as well have a bulls-eye painted on it. It’s not the place the human would want to be, yet it’s also clearly the place a human needs to be. Would it be so hard to equip it with the kinds of high resolution cameras and all-weather imaging systems used by the better-funded Grand Challenge teams? The remote driver could remain somewhere in the convoy, where he would be close enough to react to any environmental concerns unrelated to the road ahead. This would also make it unnecessary to clog communications satellites with high-bandwidth imagery the way remotely piloted surveillance aircraft do.

But driving the convoys from the comfort of a base in Tennessee would definitely have some advantages. Besides getting humans out of harms way, it reduces the logistical needs abroad, eliminating the need for some of those trucks entirely. Remote truckers don’t need their food, water, and shelter transported to the other side of the globe.

When semi-autonomous driving and remote driving catch on with the military, you can bet that civilian freight lines won’t be far behind. There will be some political hurdles, of course, which should not be understated; people won’t like the idea of potentially buggy computers driving a tractor trailer next to them at 75 mph. But the vision I have in mind doesn’t seem too far fetched.

In the near term, semi-autonomous trucks would be confined to the highways along major long-haul routes—the kinds of places where it’s not unusual even today to see two or three trucks from the same freight line keeping an inline formation. Local drivers would move long haul freight to and from waystations just off the highway where the long-haul drivers would pick up two or three trucks at a time that were going his way. These would be dropped off at the waystation nearest their destination, or at a waystation acting as a hub to other waystations. Distribution centers for major retailers like Wal-Mart are already located just off highways, often with dedicated off-ramps outside of city limits, so the biggest waystation hubs are already in place.

The robotic trucks in a convoy would be designated as such with a standard pattern of lights, reflectors, and symbol placards to help other drivers realize that these vehicles may not be as forgiving in some situations. They would also be equipped with a 360 degree array of the kinds of radar or laser rangefinders found in today's “smart” cruise control systems, providing the human convoy pilots with a clear picture of the traffic around them. At the human’s direction, or on their own, if need be, convoy trucks could be temporarily routed to different lanes, or allow non-convoy vehicles to come between them.

In the longer term, the humans in the loop would become more like ground-traffic controllers, remotely overseeing many autonomous trucks in many areas simultaneously. These operators would spend most of their time helping trucks through stretches of high traffic or ambiguous road conditions, and many would probably specialize in a few tricky locales. As a truck enters a difficult area or legally mandated safe corridor, command is routed to the available operator best equipped for it.

Ultimately, semi-autonomous and fully autonomous drivers will be safer than a fleet of human drivers, both in peace and in war. Trucks without drivers may become less tempting targets for guerillas looking for media attention, saving the lives of both drivers and bystanders. Long haul truckers will be less likely to be hypnotized by the road or fall asleep on monotonous, uneventful stretches because it is here that they will be coordinating multiple vehicles at once--and remote drivers acting collectively would never have any incentive to drive when sleepy; there would always be someone else close by who could do it. And, finally, the logical conclusion of DARPA-style projects will be artificial driver-savants that, while subhuman by many measures of intelligence, are better drivers than humans could ever hope to be.

I just don’t see them making it to mass production in time to meet that 2015 goal.


Anonymous Anonymous said...

civilian remote driven trucks are frightening. driver not onboard - driver has no incentive of personal injury to drive carefully

11:46 AM  

Post a Comment

<< Home