Presentation
by John Walker
at the IEEE Asilomar Microprocessor Workshop
April 19th, 1995
It's great to be back at Asilomar! One of the things I enjoy most about this workshop is hearing about new microprocessor applications; it's amazing that after 21 years of working with these magical slabs of silicon, we're still finding fascinating new ways to use them.
My talk today will be a little different. I'm going to describe a microprocessor application which doesn't exist, at least not yet. Now you may accuse me of crossing the frontier of vaporware into the unexplored territory of vaporsystems. I'm doing it because I believe it may be possible to find an engineering solution to what is generally viewed as a political, military, or humanitarian problem.
Solving the problem presents challenges in robot mechanics and mobility, sensors and sensor fusion, autonomous or semi-autonomous navigation, and even low-cost pyrotechnics. Yet this is a project you can build in your garage, test in your back yard, and where demonstrated success can lead to substantial funding and personal renown.
Before getting into the details, I'd like to turn the clock back 50 years, to the closing months of World War II, when the terror weapons which would shape the postwar world first emerged.
War-weary Germans learned of the development of the V-2, the first operational ballistic missile, in their movie newsreels of January 21, 1945.
No newsreel in early 1945 spoke of the preparations at the Trinity site in the New Mexico desert leading to the successful test of the first fission bomb on July 16, 1945.
The progeny of these secret weapons, the intercontinental missile and the thermonuclear warhead, would in large part define the international arena for most of our lives.
The same week Berliners learned of the V-2, recently liberated Parisians saw the first glimpse of another German secret weapon, the Bakelite plastic land mine, discovered by Allied troops advancing after the failure of the German counter-offensive in the Battle of the Bulge.
Photo:
International Red Cross
And fifty years later, that's exactly where we are today. In Angola, Cambodia, Afghanistan and dozens of other countries, people are probing the ground with knives to find and remove plastic land mines, often laid decades before in forgotten battles.
Graphic:
International Red Cross
As the threat from ballistic missiles and nuclear weapons seems to be diminishing, the land mine remains the terror weapon of choice in conflicts around the globe. Land mines have killed and maimed more people than ballistic missiles and nuclear weapons combined. They are one of the bitterest legacies of the war.
Estimates of the number of mines currently in the ground range from 85 to more than 200 million. Another 100 million are in inventory, waiting to be laid.
Most of these mines cost less than 5 dollars.
The statistics are daunting: about 100 million mines now in place in 62 countries. For every mine removed, 20 more are laid in the conflicts you see daily on CNN. Demining is a dangerous business, with one death per 5000 mines cleared, and it is a slow one. At the present rate, and assuming no new mines are laid, clearing all the mines in Afghanistan will take more than 4000 years.
Unlike other weapons, when the war is over the mines remain. Today's mines remain active for 50 to 75 years or more. Most are placed haphazardly with no record of precise location. Made almost entirely of plastic, World War II-style metal detectors are useless to find them. Casualties from mines are overwhelmingly civilians.
In Angola, where the long post-colonial war appears to be winding down at last, there are more mines in the ground than people in the country. Antipersonnel mines are not generally designed to kill, but instead to cause loss of limbs. Nick Bateman, a former British army officer puts it this way, “You kill a guy with a land mine, he's dead. End of story. But you blow his leg off, you tie up a medic for a day, and you demoralise his friends and family for years.”
There are more than 70,000 land mine amputees in Angola today, one out of every 140 people, and the number is rising at a rate of more than 2000 a year. Fewer than 10% ever receive prosthetic limbs.
Most work in mine clearance has been oriented toward combat or post-battle objectives. This is very different from post-war area clearance, and the technologies differ accordingly.
In combat, engineers work to clear a path to permit troops and vehicles to advance. Heavily-armoured tanks that flail the ground with chains have been employed for this since D-day--they simply ride out the explosions, and every now and then you replace the chains.
After the battle, the priority is de-mining roads to permit resupply. The path is narrow and well-defined, and the terrain is amenable to wheeled and tracked vehicles. Other mined areas are dealt with simply by posting a “Danger—Mines” sign.
It's a very different matter to de-mine broad areas of land after a protracted and chaotic conflict where, in many cases, mines were deliberately laid to terrorise and control civilians and deprive them of the use of agricultural land.
You have to cover a wide area, including all kinds of terrain, and you need a high confidence all the mines have been neutralised before sending your kids out to work in the fields.
As I'm sure survivors of my earlier talks here will recall, I have a tendency to believe most problems are, at the root, economic. I certainly believe that's the case here. As long as it costs about 3 dollars to lay a mine and on the order of a thousand dollars to find it and dig it up, the mine-layer has an overwhelming advantage over the mine-remover. As long as these fundamentals remain in place, the number of mines in the ground will continue to grow.
Arms control initiatives can be viewed as attempts to raise the cost of laying a mine by restricting supply. Nobody believes it's possible to substantially close the gap that way.
So, we're faced with what appears to be an insuperable economic problem. If we were lawyers or politicians, we'd just give up but, hey, we're engineers. Let's hack.
Maybe we can fix this! After all, a three-hundred-to-one price reduction is nothing compared to what we've lived through over the last couple of decades. Is there a way we can apply the microprocessor to clearing land mines and, by doing so, couple the inexorable engine of Moore's law—the ongoing increase in processing power at constant cost, to drive down the removal cost and, as it were, level the minefield by narrowing the cost gap between laying a mine and getting rid of it.
It's amazing what you can accomplish with clever engineering riding the back of a multi-decade technological trend. Ten years ago were we marveling, “Memory is free.” Might we, in a few years, say the same thing about mine clearance?
Anyway, that's what I've been thinking about off and on for the last couple of years. Most current military mine clearance equipment and civilian bomb disposal gear is technologically akin to a 370/168 mainframe: big, heavy, expensive, and highly specialised for its mission.
Why not explore the other end of the scale? Smaller, cheaper, and smarter seems to work awfully well in a variety of situations. Why not in the minefield? In particular, might it be possible to build a small, light, inexpensive robot which could find buried mines and, in some manner, dispose of them?
Instead of a huge tank rumbling around, occasionally detonating a mine, why not a bunch of rat robots sniffing out the mines somehow? Rats and rabbits don't set off mines—if they did, the mine problem would quickly take care of itself.
Minerats!
The more I thought about it, the more I became convinced it just might be possible.
In any case, this looked like a research problem that could drive mobile robot technology in useful directions. You sometimes get the impression that you have to go to the Moon or Mars or at least a volcano in Antarctica to find something useful for a robot to do.
But here's a job crying out for cheap, lightweight, highly mobile, all-terrain robots which, if you managed to build them, could immediately be put to the test and set to work all around the globe with an immediate humanitarian payoff. They could be tested without expensive environment simulators, deployed without enormous launch costs, and refined in the field with the quick turn-around effective engineering requires.
Feedback from the environment is essential to rapid evolution. Sudden, violent death is a highly effective and time-proven form of negative feedback.
Here's what seemed to make sense for low-cost mine clearance. I'm not a great believer in a priori design; any of these points may change during development or based upon field experience.
There's no reason to go for full autonomy—we're not half a light-hour away on Mars, just hunkered down far enough away to be safe if something goes boom. Besides, full autonomy in real-world terrain is wildly beyond the state of the art.
We should be able to make a system, though, that can be operated by local talent with no more training than it takes to play a Nintendo game. Perhaps we can enlist a game designer to make it as addictive—Super Mario Minerat. Well, probably not.
Cleared terrain, especially agricultural land, has immediate economic value. This translates into compensation for the operator.
Fully manual operation, like a remote-controlled lawnmower, isn't desirable, since we need a high level of confidence that all mines have been detected. We need verification of a complete sensor sweep of the area, and that implies enough navigation capability to delegate some of the responsibility to the robot. Besides, automating wherever possible reduces the risk of errors due to boredom in a human operator.
In mine clearance, weight (or more accurately static ground pressure and impact force) is a step function, not a continuum. Below the threshold that explodes a mine (except for rare rogue or defective ones), you don't have to worry about avoiding them. Above the limit, you either need to detect, with a high level of confidence, every mine in time to avoid it, or else be so heavily armoured you can survive detonations without major damage.
Rabbits mark the division between light- and heavyweight approaches. If rabbits set off mines, most mines would be quickly exploded. Population doubling time in rabbits is such that loss to mines would be a minor cause of death. So, as long as we have less impact on the terrain than a rabbit, we can dispense with heavy armour. We don't even need real-time sensor analysis; this may make some approaches, like synthetic aperture radar, feasible.
However challenging it may be to build, the robot is nothing more than a platform to carry the sensors. The high-tech brain is the microprocessor and sensor package, and it's here most of the high-powered engineering is needed. Given the crude standards of mine construction, the clearance robots may trigger an occasional explosion. Epoxy-potting the electronics, perhaps with rudimentary armour, should enable it to be reused after these misadventures. On the other hand, we may decide it's cheap enough to consider expendable based on on the actual rate of loss.
The design should use only components available worldwide and free of all export restrictions. Notwithstanding policies of certain countries, I don't believe Iraqi and Iranian kids should have their legs blown off because their governments happen to be out of favour.
We should certainly feel free to bet on ever-increasing processor power. A molasses-slow compute-bound design will probably run as fast as the mechanics permit by the time it's in wide use in the field.
The mechanical design should be as simple and low-tech as we can make it. I suggest we strive for maximum parts commonality with bicycles: you can get a bicycle fixed almost anywhere on Earth. A Minerat based on bike technology should be able to be locally fabricated and repaired when necessary.
Mountain bikes demonstrate, at least in the hands of lunatic riders, the ability to negotiate a broad variety of terrain. Their components may be applicable to our slower, more risk-averse robots.
I'm not a Northern chauvinist. On the contrary, I'm trying to be sensitive to the fact that many apparent First World constants aren't invariant under latitude inversion.
In the First World, agricultural land is not highly valued; over the last 50 years, despite population growth, marginal farmland in Europe has reverted to forest. But within the lifetime of the younger people in this room, the world's population will double from 5 to 10 billion, and most of those new people will be in the Third World. It's too late to stop it or slow it—the parents of the 10 billion are already maturing. Human dignity and survival in this world will require an agriculture, both intensive and extensive, surpassing anything in the human experience. That means lots of arable land, and lots of that land is mined and unusable today.
When we budget, we often assume salaries are the largest component. Twelve hours from now, at dawn in Angola, people will walk into the fields to clear mines by hand. They earn 50 dollars a month doing this dangerous work. That is ten times the salary of teacher.
Knowledge of and access to high-technology components are almost nonexistent. We need to encapsulate the high tech stuff into a box that a bicycle mechanic can turn into a Minerat.
Local materials, local manufacturing, and local manpower are essential to making this work. We must design with that in mind.
Finally, we need a cheap way to explode detected mines in place. The alternatives are using separate removal robots or sending humans into the minefield to dig up detected mines; cost and risk to life preclude these.
Better to blow them where they lie. Explosions are entirely acceptable in mined terrain, and provide unambiguous confirmation of destruction.
We should be able to develop a shotgun shell-sized shaped charge, costing much less than one dollar in million lots, which will detonate a mine in place.
The cheaper we make the charges, the more false positives we can tolerate. That reduces the performance criterion for the sensor suite and detection capability. We should let economics decide this—90% false detections may be the optimal point, and we shouldn't be offended if it works out that way.
When you examine mine clearance as a systems integration problem, it quickly becomes apparent that the sensor is the crucial component. If we can't detect mines in the ground, or if the only feasible means are heavy and expensive, like million dollar thermal neutron activation machines used for airport baggage screening, it doesn't matter whether or not we solve any of the other problems.
Conversely, if we find a cheap, lightweight, and reliable sensor and discover we can't build a suitable robot, we could sweep it over the ground at the end of a long boom.
Several kinds of sensors appear promising. Plastic mines don't entirely contain the characteristic odour of nitrate-based explosives. Trained dogs have demonstrated the ability to detect buried mines, even metal mines in many cases. Dogs are impractical for area clearance due to the extensive training, lack of attention span, and the need for a human handler, but they do show that mine detection though vapor emission is feasible. Development of micro-machined silicon sensors capable of detecting picowatt exothermic reactions may, in conjunction with a suitable catalyst, yield effective sensors at chip prices.
Monopulse or ultrawideband radar has demonstrated the ability to image more than one metre into most soil, plenty deep enough to detect mines. Lawrence Livermore has licensed such technology and it is currently being developed for commercial applications as mundane as stud finders.
Synthetic aperture radar has also demonstrated adequate ground-penetrating capability. While existing systems are large, it may be possible to develop a hybrid system with a large transmitting antenna and individual local receiving antennas on the robots.
This is obviously a project still in the formative stage. Solving the land mine problem is going to require knowledge in a wide variety of fields and clever real-world engineering. If you have expertise or know people who do, let's talk off-line.
Talking about vaporsystems isn't nearly as satisfying as demoing hardware. In a couple of years, maybe we'll be able to stage another Asilomar first—the first-ever All-Robot Exploding Easter Egg hunt. It'd be great if three or four entries were inspired by this talk.
Thanks.