Telegrasping at Midnight ‘Neath the Starry Sky


Even though written almost a decade after the demise of the Environmental Ecology Lab, this piece provides some useful information about the initial design of its Telegrasp technology and the role that self-organizing controllers were to play in it.

Think of it. An age-old dream as recurrent as the hunger for unaided flight, a communication scheme so rich in its complexity that one could actually make love to someone else at great distances of space (or maybe even time? Casanova, Marilyn Mon­roe, the Tantric Sages?). Dream on, friend, it’s not in the cards, although occasionally our technological hubris has led us to think it all might be wait­ing right around the corner.

Shepherd Mead of How to Succeed in Business... fame wrote a hilarious satire in the ’50s called The Big Ball of Wax in which love and religions were marketed vicariously with a helmet communicator. Many other sci-fi au­thors have proposed their variations. And then in 1969 a hoax called “Intersex” was announced in the Sep­tember issue of Architectural Design as a news item that would have done credit to Orson Welles —but again, no substance to it.

Now, I don’t want to spoil anyone’s fun in speculating on these matters, but the problem boils down funda­mentally to the impossibility of sepa­rating out what is input from what is output in the richness of intimate dia­logue. Science and technology have thrived so grandly in all those places where observer and observed could be distinguished that it’s both a shame and a relief to be able to wag a finger and say, “Sorry, gotcha this time!” Touch and movement —the haptic reflex loop —are such that that which touches also moves and is itself touched, and there is nowhere between lovers that you can pass your mathematical plane and say, “Here, see? This message is going this way and that one the other way.” No way. No surrogate simulation can reproduce all the wonderments of intimacy, no matter how many sensors and trans­ducers you care to incorporate at the interface: the information shared by two real people is just not of that nature.

You might be able to build a clockwork doll of modern hi-tech to play an exciting game with you —for I’ll never discount the ability of an in­dividual to fantasize the rest. That’s not what I’m addressing here. What can we do that might be worth doing to enrich, let’s say, the intimacy of a telephone call? Is there a way to share touch and movement at a distance, a way of providing our remote out-reaching with context we normally think of as exclusively the domain of face-to-face presence? Consider TELEGRASP. Maybe it’s an idea whose time has come.

What’s really involved in what we call “holding hands”? Is there some­thing special about hands and their modes of grasping that is unique? I don’t think so. For the kind of contact I have in mind, an arm around the shoulder, knees touching beneath a table, the nudge of an elbow, or any sort of labile pressure can make a mes­sage spoken very differently received by the hearer being touched from what it means for someone else, and that’s the point: we’re looking for a simple exchange of responsive touching that can make a difference in a dialogue. If we resort to thinking about surrogate hands and the making of measure­ments upon them as to how they are being touched, or the making of simi­lar measurements on the real hands of the respondents, we will again have lost the game. It turns out to be a lot easier than that.

Imagine that we have provided two environments for hands, one at each end of the line. They might look like enclosures the size of a shoebox, with an opening at one end. Say that they each contain about half a dozen in­flatable bags that would more than fill the space if all were inflated fully. I insert my hand in one, you yours in the other, and the bags inflate to enclose our hands everywhere with a firm pressure.

Contained with the bags are about an equal number of sensors engaged in making some quantitative measures on one or more bags: some might be cir­cumferential strain gauges, others might sense the extent of contact area between two bags, and another per­haps would monitor the net rate of air flow into and out of all the bags taken as a group. The point to notice is that the system is making its measurements upon itself, not upon your hand or mine. If the two environments have been made reasonably identical in size and placement of bags and of the sen­sory apparatus, then our hands may be thought of as intrusions into two identical spaces, and the differences in the information retrieved from the respective sensors will give some meas­ure of what our two hands are doing differently.

Let us now suppose that the differ­ences, by respective pairs, are introduced as a set of error signals to a “multiple-goal, multiple-actuator, self-organizing controller.” Well, now, you ask, “What the hell is that?” I don’t really want to explain fully here what an MGMASOC is, but I’ll sum­marize. Back in the ’60s, when some pretty hairy control problems present­ed themselves to the designers of supersonic aircraft and space satellites, the control-systems people came up with controllers that could do some as­tounding jobs very well. You could give the controller information about how well it was doing moment-to-moment relative to achieving a hand­ful of goals (the systems people called this “performance assessment”), and at the same time you gave the control­ler carte blanche to fiddle with every available output to the system (the “actuators”), even though some of its decisions in the short run looked crazy —oh, excuse me, “counter­intuitive” sounds better.

Anyway, these controllers can keep a system on track so well—even in the face of disturbances like variables sud­denly reversing the meaning of up and down or some part of the system failing altogether - that no sophisticated “de­cision theory” structures can keep up with them. Nowadays the SOCs are controlling things like steel rolling mills, but there’s an unlimited assort­ment of potential applications like toys and greenhouse environmental con­trols that can use them to good advan­tage. I think it won’t be long before SOC-on-a-chip is with us.

In our case—TELEGRASP—the self-organizing controller will look at the paired measurements from the hand boxes as error signals to be nulli­fied, and it will have control of the air bags to try to accomplish the task. We will make only the simplifying stipu­lation that the same command. “Inflate” or “Deflate,” must go simul­taneously to the respective bags at the two ends of the line. What will happen?

Unless the two of us remain terribly flaccid and unresponsive, the system will engage in a constant struggle to bring about a sensed condition of identity between the two ends. The goal is unattainable, relatively speak­ing. but the struggle will be felt by both of us. The changes in the struggle that result from my movements will be redundant for me and will go essen­tially unnoticed, but those that result from your movements will charge my handbox with something of your presence and, as adjuncts to our con­versation, will put us literally more “in touch with” each other’s feelings and reactions.

Anybody want to try it? (Try building TELEGRASP, that is.) The realization of the system as a laborato­ry toy seems simple enough, and as a telephone add-on it doesn’t present much of a bandwidth problem. I made a start on it back in 1968, but our lab­oratory folded and I went on to other things. To my knowledge, the name TELEGRASP has never been copy­righted, nor has an embodiment of the system been patented. I’d enjoy seeing the gadget emerge some day. 


As an illustration of contrasting approaches to a control problem — classical “analytical” method versus self-organizing controller (SOC) — consider how one might aim an orbiting reconnaissance satellite. The problem is as follows: At in­tervals you want to reposition the satellite’s axis with respect to the earth’s surface, but since the vehicle is to be in orbit for years, you can’t afford to send along enough thruster fuel to last. However, solar panels can absorb and store electric energy indefinitely, there is a mag­netic field around the earth, and a current passed through a coil can elicit a torque from that field. So your vehicle carries three coils, placed at right angles to one another, through which currents may be passed for aiming the cameras. Now the question that remains is: how much current in which coils and when? Remember also that you want to conserve energy resources (current) when­ever possible.

The classical approach directs you to have an onboard device measure the strength and direction of the earth’s magnetic field in the place where the satellite is now; other sensors are to report on the attitude of the satellite with respect to the limb of the earth. All this in­formation is sent to a computer on the earth’s surface, and computa­tion proceeds.

The first step is usually to make an estimate of what the magnetic field around the satellite will be like when the entire computation is completed. That alone is no trivial task —nor can it really be carried out precisely—because of the gross irregularities in the magnetic field in some areas. Next are solved the geometric and energetics equations that will make possible the rotations required, given the available re­sources, and these are usually worked out one axis of rotation at a time. Then the whole procedure has to be carried out a few more times to refine the positioning further, since the first adjustment is only an approximation. The meth­od is time-consuming and energy­wasteful, and it will not account for changes in the functioning of the satellite’s sensors or coil “actua­tors.” Implementation is difficult and fraught with sources of error.

Let us instead put an SOC on board, give it control over sending currents through coils, give it access to how well it is doing relative to its goal, that is, give it the moment-to-moment information about the satellite’s orientation relative to the limb of the earth. The command from earth is “Go” and it goes from where it is to where it ought to be —in one apparently continuous movement such as you might make to reach for a pencil on a shelf be­hind you, without having to command each of the dozens of muscles separately involved in the task. The SOC does not have to know a priori which current direction is “up” for which coil; it finds out very rapidly by perfor­ming random experiments in mi­crotime (a very short sampling interval relative to the response interval of the vehicle). In fact, it doesn’t have to know that there is a magnetic field out there to interact with, nor what size or direction it might have at the moment. And furthermore (this is for all you armchair physicists who think you really know what is “out there” in space), if there were somehow another field present that currents in coils could push on, the SOC would make immediate and rele­vant use of it; the measure- compute-decide-command system in the “analytical” example might never know this other field was there and fail hopelessly — needlessly — in its task.

A subgoal, perhaps of lower priority than the accurate and rapid positioning of the satellite, can be the conservation of energy by the entire system, and this subgoal will be cultivated commen­surately with its priority. If some part of the system starts to fail, the SOC can “work around it.” The performance will be slower, per­haps, or of lower resolution, but the job will still be done as well as possible. The aerospace industry calls this “graceful degradation.” Nice. In the classical case, if you don’t tell the computer on the ground that something has changed, it will give commands as if everything is up to par —and err accordingly.

The magic of the SOC lies in its ability to perform its job without a priori knowledge of the connec­tedness of its environment. But the internal organization is astoundingly simple. Random-noise generators, low-resolution means of keeping track of successes in the recent past, and a rich cross-connection of control pathways are the essential ingredients. Nothing precise. Loose and flexible. SOCs will be most useful where natural growth systems are enclosed in artificial containers: Greenhouses, aquariums, (schools?), prostheses. The SOC produces behavior best described as heed without habit.


Ed., the following is inlaid midway through the article.