Analog Robots

Dateline: June 7, 1998

The "traditional" robot has a computer brain. Because computers are digital, we therefore tend to think of robots controlled by them as being digital also, although the bulk of their bodies are actually made of analog electro-mechanical components.

Analog robots are brainless. They are controlled not by some program stored in a memory bank, but by natural algorithms inherent to their analog components.

For a simple example: a thermostat is an analog device. The simplest consists of a bar made up of two different metals each of which expands at a different rate when heated (and contract at different rates when cooled). The metals are glued together along their lengths, and will bend one way then back the other as heat is added and removed respectively. The algorithm—IF it gets warmer THEN expand; IF it gets cooler THEN contract; ELSE stay the same—is a natural corollary to a law of physics; it does not have to be programmed in.

Scientists at Los Alamos National Laboratory, birthplace of the atom bomb, have been tinkering with all sorts of analog electromechanical devices—capacitors, resistors, diodes, gears, wheels, lenses, rods—stripped from junked computers, VCRs, CD and record players (remember them?), cameras, telephones and pagers, oscilloscopes, TVs, and so on, and connecting them together to see what natural algorithms would lead them to do.

And lo and behold, when strung together in some configurations, they behave like living things. Not (yet) very advanced living things; insects, mainly.

The Los Alamos boffins are not simply having fun making toys at taxpayer expense (or if they are, it is incidental). There's a serious purpose behind the work, as you can see from this quote from the LANL Web site:

Dependable machines capable of working in unstructured, unpredictable environments can readily conduct a number of operations, including monitoring, security, cleanup, and maintenance. These machines can work without rest in environments hazardous to humans. These new types of machines consist of a patented, new control paradigm for robotic control called Nervous Nets, which is based on minimal, elegant, non-linear, analog electromechanics. Keyed on fundamental principles of artificial life, BIOMORPH robots exhibit characteristics that allow them to solve real-world problems in a biological manner—in essence, to adapt to an ever-changing environment. Adaptive robot swarms can be configured to carry sensors or tools under local or remote operations. The applications for analog robots envisaged by LANL include environmental management, adaptive mechanics, exploration and monitoring, security and maintenance, and "advanced AI platforms." Among the advantages LANL lists for analog robots are their "self-programming" ability (which I take to mean their natural tendency to "learn") and their longevity (they can learn to feed themselves—to seek an energy supply when their batteries run low).

LANL also says that they are "compliant, reconfigurable, energy efficient, biodegradable, and inexpensive." Having already created 70 engineering prototypes and with 30 prototypes under active development, LANL should know!

In the context of analog computing, which we discussed in last week's feature, LANL's "Nervous Net" control model is very interesting. As you may recall, some organizations are now producing analog VLSI chips specifically for neural net applications. If such chips were implanted in analog robots and trained to LANL's Nervous Net paradigm, then we might see a new breed of brainy analog robots, and it is very conceivable that they might be smarter than their digitally-controlled cousins.

If you are among those who believe that true AI—a self-conscious artificial lifeform with a high degree of intelligence (at least equivalent to human intelligence)—is not possible to emulate digitally, then maybe analog robots plus analog computing will do the trick.
 

  Until next week, 

 

NEXT WEEK: Some brief book reviews; odds and ends.

Help Wanted: Got questions or comments on this article or on any other AI-related subject under the sun? Post it in the AIBB!

Previous Features