Essays. FSF Columns

"Robotica '93"

We are now seven years away from the twenty-first century. Where are all our robots?

A faithful reader of SF from the 1940s and '50s might be surprised to learn that we're not hip-deep in robots by now. By this time, robots ought to be making our breakfasts, fetching our newspapers, and driving our atomic-powered personal helicopters. But this has not come to pass, and the reason is simple.

We don't have any robot brains.

The challenge of independent movement and real-time perception in a natural environment has simply proved too daunting for robot technology. We can build pieces of robots in plenty. We have thousands of robot arms in 1993. We have workable robot wheels and even a few workable robot legs. We have workable sensors for robots and plenty of popular, industrial, academic and military interest in robotics. But a workable robot brain remains beyond us.

For decades, the core of artificial-intelligence research has involved programming machines to build elaborate symbolic representations of the world. Those symbols are then manipulated, in the hope that this will lead to a mechanical comprehension of reality that can match the performance of organic brains.

Success here has been very limited. In the glorious early days of AI research, it seemed likely that if a machine could be taught to play chess at grandmaster level, then a "simple" task like making breakfast would be a snap. Alas, we now know that advanced reasoning skills have very little to do with everyday achievements such as walking, seeing, touching and listening. If humans had to "reason out" the process of getting up and walking out the front door through subroutines and logical deduction, then we'd never budge from the couch. These are things we humans do "automatically," but that doesn't make them easy -- they only seem easy to us because we're organic. For a robot, "advanced" achievements of the human brain, such as logic and mathematical skill, are relatively easy to mimic. But skills that even a mouse can manage brilliantly are daunting in the extreme for machines.

In 1993, we have thousands of machines that we commonly call "robots." We have robot manufacturing companies and national and international robot trade associations. But in all honesty, those robots of 1993 scarcely deserve the name. The term "robot" was invented in 1921 by the Czech playwright Karel Capek, for a stage drama. The word "robot" came from the Czech term for "drudge" or "serf." Capek's imaginary robots were made of manufactured artificial flesh, not metal, and were very humanlike, so much so that they could actually have sex and reproduce (after exterminating the humans that created them). Capek's "robots" would probably be called "androids" today, but they established the general concept for robots: a humanoid machine.

If you look up the term "robot" in a modern dictionary, you'll find that "robots" are supposed to be machines that resemble human beings and do mechanical, routine tasks in response to commands.

Robots of this classic sort are vanishingly scarce in 1993. We simply don't have any proper brains for them, and they can scarcely venture far off the drawing board without falling all over themselves. We do, however, have enormous numbers of mechanical robot arms in daily use today. The robot industry in 1993 is mostly in the business of retailing robot arms.

There's a rather narrow range in modern industry for robot arms. The commercial niche for robotics is menaced by cheap human manual labor on one side and by so-called "hard automation" on the other. This niche may be narrow, but it's nevertheless very real; in the US alone, it's worth about 500 million dollars a year. Over the past thirty years, a lot of useful technological lessons have been learned in the iron-arms industry.

Japan today possesses over sixty percent of the entire world population in robots. Japanese industry won this success by successfully ignoring much of the glamorized rhetoric of classic robots and concentrating on actual workaday industrial uses for a brainless robot arm. European and American manufacturers, by contrast, built overly complex, multi-purpose, sophisticated arms with advanced controllers and reams of high-level programming code. As a result, their reliability was poor, and in the grueling environment of the assembly line, they frequently broke down. Japanese robots were less like the SF concept of robots, and therefore flourished rather better in the real world. The simpler Japanese robots were highly reliable, low in cost, and quick to repay their investment.

Although Americans own many of the basic patents in robotics, today there are no major American robot manufacturers. American robotics concentrates on narrow, ultra-high-tech, specialized applications and, of course, military applications. The robot population in the United States in 1992 was about 40,000, most of them in automobile manufacturing. Japan by contrast has a whopping 275,000 robots (more or less, depending on the definition). Every First World economy has at least some machines they can proudly call robots; Germany about 30,000, Italy 9,000 or so, France around 13,000, Britain 8,000 and so forth. Surprisingly, there are large numbers in Poland and China.

Robot arms have not grown much smarter over the years. Making them smarter has so far proved to be commercially counterproductive. Instead, robot arms have become much better at their primary abilities: repetition and accuracy. Repetition and accuracy are the real selling- points in the robot arm biz. A robot arm was once considered a thing of loveliness if it could reliably shove products around to within a tenth of an inch or so. Today, however, robots have moved into microchip assembly, and many are now fantastically accurate. IBM's "fine positioner," for instance, has a gripper that floats on a thin layer of compressed air and moves in response to computer-controlled electromagnetic fields. It has an accuracy of two tenths of a micron. One micron is one millionth of a meter. On this scale, grains of dust loom like monstrous boulders.

CBW Automation's T-190 model arm is not only accurate, but wickedly fast. This arm plucks castings from hot molds in less than a tenth of a second, repeatedly whipping the products back and forth from 0 to 30 miles per hour in half the time it takes to blink.

Despite these impressive achievements, however, most conventional robot arms in 1993 have very pronounced limits. Few robot arms can move a load heavier than 10 kilograms without severe problems in accuracy. The links and joints within the arm flex in ways difficult to predict, especially as wear begins to mount. Of course it's possible to stiffen the arm with reinforcements, but then the arm itself becomes ungainly and full of unpredictable inertia. Worse yet, the energy required to move a heavier arm adds to manufacturing costs. Thanks to this surprising flimsiness in a machine's metal arm, the major applications for industrial robots today are welding, spraying, coating, sealing, and gluing. These are activities that involve a light and steady movement of relatively small amounts of material.

Robots thrive in the conditions known in the industry as "The 3 D's": Dirty, Dull, and Dangerous. If it's too hot, too cold, too dark, too cramped, or, best of all, if it's toxic and/or smells really bad, then a robot may well be just your man for the job!

When it comes to Dirty, Dull and Dangerous, few groups in the world can rival the military. It's natural therefore that military-industrial companies such as Grumman, Martin Marietta and Westinghouse are extensively involved in modern military-robotics. Robot weaponry and robot surveillance fit in well with modern US military tactical theory, which emphasizes "force multipliers" to reduce US combat casualties and offset the relative US weakness in raw manpower.