Definitions, Examples, and Ideas [Guide to Advanced Robotics]

What is robotics? This is not a new word but it may be an unfamiliar one. Automation is another somewhat similar word that applies in many instances to the same idea or concept. It is only through the magic of the big screen TV and science fiction stories that the word robot has come into such prominence. It is now almost a household word. The word robot has come to be associated with the C3POs of other worlds; those mechanical, man-like units of metal and electronics that play roles in stories of the future. It is interesting that the word robot can also mean an unthinking robot being who performs some task automatically. It is not considered normal for a human because that is the way machines are supposed to operate.

Machines act and do not think--or at least do not think or reason as humans think and reason.

Robots do have memory banks that are usually programmed in some manner. They do speak and seem able to understand human speech. Because we have so many varieties of solid-state circuits nowadays, no one is surprised when a robot speaks, or a calculator, game, TV or automobile speaks in a distinct monotone, but remember, they don't think!

The word robot comes from the Czechoslovakian word robotnik meaning slave, servant, or compulsory service. The word has been defined as being descriptive of an automaton or a person who acts or works mechanically without thinking for himself. By definition, a robot pilot is a device which serves as an automatic pilot in an airplane for example. Note that the word automatic appears in the definition.

During World War II the concept of an automatic pilot was extended to guided missiles (also called robot bombs). It kept them flying upright and capable of executing steering commands sent from an automatic ground computer controller. This points to another link in the system-the data or command transmission link.

The exploration of space has provided robots of a much more sophisticated variety. These mechanical marvels seem to be able to think and act on their own as they travel into the vastness of the cosmos. These robots have the ability to sense situations and conditions. Automatically they extend measuring or photographic instruments from protective housings and sample the ether, or planets, or asteroids. They place the information in memory banks and then send it, slowly, allowing plenty of time for accurate transmission, back to the controller-computer on earth where the data is assembled, enhanced, and displayed for human consideration and amazement. When so programmed, these robot marvels steer themselves down to a planet's surface. They land care fully, adjust themselves, and communicate their status, position, and findings, in this new environment. It has been rumored that the Russians have developed this type robot to such a state of perfection that they land on the moon, take samples, lift off, and return to earth without the assistance of human control in flight or during on-planet actions. They act like the mechanical men of Karel Copek's play R. U.R. even if they do not have humanoid appearance.

There can be no doubt that man is moving toward that day when some type of machine will be used in almost every human capacity. The scientists believe that people would like to have this machine look somewhat human, and so they spend considerable time in an effort to develop mechanical eyes, ears, vocal chords, arms, hands, legs, and a method of locomotion which simulates human walking. Everyone is de lighted when a mechanical machine, looking very much like a Middle Ages knight-in-armor clumps around and performs some act of a semi-useful nature, or utters sounds which are human words, all in proper response to voiced human commands.

While it all seems relatively normal, response of a machine to human commands uttered vocally is not simple. It is the world of algorithms which is what the scientists and engineers call their magic equations. Algorithms instruct and control the machine and everything it does. Associated with every type of mechanical device which does anything, there is some type of controlling device which reads an algorithm. In order to understand and communicate directly, we should also have some understanding of algorithms which are used in both operational and developmental situations to make things come alive.

When we consider all the parts of the world of robotics, we are talking about robotics. In the Introduction

I have described this as being the art or science of controlling machines such that they do something. We now consider what that something might include and thus consider in more detail what we might mean by saying that robotics is an art.

Art is defined as a skill acquired by experience or study.

It also has been defined as the systematic use of knowledge or skill in making or doing things, and the use of skill and imagination in the production of things of beauty, or the things produced which are so considered. So it seems from these definitions that robotics means machines, computers, mathematics, algorithms, science, engineering, and some artistry. If we try to come up with a simple kind of definition for robotics we might try: The design, use, and operation of machines, which are computer controlled by algorithm, to do human-desired tasks.

This definition does not include any specification as to the shape or size or type of machine, nor does it limit itself to a definition of a physical output-although that kind of output is usually associated with the word robot. Notice also that we have indicated the machine must be controlled by an algorithm and implied in this definition is some type of computer. We try not to limit the machine to either fixed or mobile construction, and we do not limit the machine to an autonomous operating condition. You may take the extreme position that the human brain is also a computer! This means that waldos are also robots.

The incorporation of a human's brain somewhere within the flow pattern of operational commands for a mechanized unit has to be considered at some stage, even though it may be by-passed at later stages and in subsequent operations of the electromechanically directed unit. If you ask what this means, I would respond, "We are referring to the teaching stage in development of a system which will then become autonomous.

"What about self learning?" you might ask. That comes under the heading of adaptive control, which simply means that the machine can adapt itself to changing conditions of input that determine its operational output. In this case the machine is equipped with sensors that can recognize changes in the position of objects that it is to pick up and convey to another place. As it determines these changes, its sensor input can cause changes or modifications to its control computer program so that the machine's mechanical movements can be modified to adjust its pick up arm or hand or gripper and continue to do its task without fault. Adaptive control capability is still controlled by the original algorithm, though.

The ability of a machine to do things of a constructive nature depends upon the accuracy, sensitivity, number, and type of it's sensors. Consider for a moment the human ability to pick up a pencil. The sensors involved would be the eyes, which determine at every instant the relative position of the hand, arm, and fingers to the pencil, and the skin sensors, which feel the pencil when it is touched. The sense of feel allows the brain to determine the tightness of grip and position of the grip necessary to pick up the pencil. The eyes help the brain determine where to move the pencil. If the underlying command of operation was "Pick up the pencil and write 50 on a paper" the eyes would also have determined the position of the paper, the skin sensors would have determined that the paper and pencil met under the proper circumstances, and then the algorithm causing the fingers to make the movements to write 50 would have been initiated.

What the human machine is getting throughout this operation is feedback from all the various sensors necessary for this kind of operation. Sensors provide input signals of what the hand is doing. What the hand is doing is the machine's output. When a machine inputs signals from its own output this is called feedback, and this is vital to the operation of any machine considered under the heading of robotics. Feedback is a concept long known in engineering circles and in the world of servomechanisms. What is new in the world of robotics is the type, size, and accuracy of the sensors, and the ability to compute very complex equations very quickly to determine exact operational signals for the robot.

The human body is one big mass of sensors. Almost every point on the skin's surface is sensitive to heat and pressure in ranges from the smallest puff of wind to the largest rib-smashing blow, and from a bitter winter morning to a humid summer scorcher. Add to this the hearing sensors, the visual sensors, and the taste and smell sensors and you have a pretty complex and sensitive package. To enrich your thinking along this line, consider that pencil picked up a few paragraphs ago. How does the brain know that it is a pencil when the eye sensors see it? If you answer that the brain has a stored image of a long slender multisided object with a sharp point, then you are getting close to the kind of thinking used to define algorithms in robotics. Think of how many images our brains must have in storage on a relatively permanent retention basis, and how it discards old images and incorporates new images. Then consider this thought provoking definition: Robotics is the science of making machines man-like in action and operation.

- To discuss a robot is one thing, to examine it is another.

Let's take a look at one kind of robot. The Grivet Series 5 industrial robot in Fig. 1 might be a surprise because it consists of a single automated arm that performs under the direction of a control unit, the TARC (Fig. 2). The Grivet responds to the teaching instructions of the person holding the teaching box in Fig. 1. We are assured that teaching this kind of robot to do important industrial tasks is easy. In fact, the American Robot Corp. says that this unit is designed to be extremely simple to use and to maintain. What about maintenance? Robotic devices are electro mechanical machines and thus their ability to function and to keep functioning for long unattended periods of time depends on how the designer planned for maintenance-and possibly troubleshooting, or diagnostics.

I am reminded of a situation concerning the development of a new guided missile robot system. Many manufacturers and scientists were present and deeply involved in the planning for the system. When it came time to consider maintenance and repair, everyone agreed that the system would "almost never" break down or need such attention. Just in case it should, a testing unit was to be devised which would automatically run tests and isolate troubles so quick and effective repairs could be made. You guessed it! The complexity of the testing unit was of such magnitude that it became advisable to consider a testing unit to test this testing unit! We want our robot systems to be as free from troubles and maintenance as possible, but what testing and maintenance they have should be easy to perform, consistent in application, and totally effective in operation.

Look back at Fig. 1. This Grivet robot has been de signed for industrial applications where a job handling objects weighing less than three pounds is the primary concern. It can be taught what to do for a particular task with just a few minutes of instruction, using the hand held teaching unit. A close-up of this unit is shown in Fig. 3. Simplicity may be a key to easy maintenance.

Fig. 1

Fig. 2.

Fig. 3. The teaching unit (pendant) for the Grivet robot (courtesy American Robot Corp.).

TEACHING A ROBOT

Robot Master, John Galaher of the American Robot Corp., informs us that this box controls extension and contraction of the arm and rotation of the hand gripper, wrist, elbow, and shoulder. One sets this kind of robot into its permanent location, programs it to accomplish the task desired, and then removes the teaching unit. The arm, under direction of TARC, will now go to work doing what you have told it to do.

In this case it may replace a person on a batch separation line, or change objects from one conveyor belt to another. It is man-like in that it can do the kind of job that a man did previously.

"Robotics is the science of making machines man-like in action and operation." This seems to be an accurate definition and you will note that it does not say that the machine has to look man-like (or woman-like) in appearance, although it doesn't rule it out either!

EMULATING HUMAN ACTIONS

In many discussions of the capabilities of a robot arm such as the one in Fig. 1, it is stated the arm will emulate human actions. Perhaps you are familiar with this word. It seems an important one.

"To emulate" means to strive to excel, or to exceed, or to equal. In this case the arm emulates the task which was formerly done by a human. Of course we have to consider the possibility that the task will be a new one, or one in a new environment or location which has never been performed by anyone, human or not. We imagine that production will re quire a whole list of tasks, resulting in a final product or service. We look at the performance of these tasks as an example, and we have to make some comparisons, so we naturally make the comparison between the machine accomplishment and a human accomplishment.

GRIVET-ROBOT ARM DYNAMICS

Let's examine that Grivet robot arm in more detail using Fig. 4 to guide our thinking.

Fig. 4. The statics and dynamics of the Grivet robot (courtesy American Robot Corp.).

In A we see the static condition of the arm. It is tightly positioned against its base, and it does nothing. In B it has been sent signals to extend, and from the illustration we can see through what angles rotary motion can be accomplished, and also via these same angles a translatory motion is made for the gripper. This is easy to understand if you imagine that section 1 of the arm moves or rotates downward, while section 2 moves or rotates back, and section 3, the wrist, rotates downward to remain on a level line from the starting position. Notice also that the arm itself can rotate through 360 degrees around the base pedestal so that the gripper can move an object from here to there, in a sideways motion.

How can it be said that this mechanical arm emulates a human action? Let's take the situation where the arm, by definition, is to excel a human capability; do it better than a human can. To begin, assume the arm and the human are equal in speed, accuracy, and general operation. Time passes, the human tires, the machine does not. Now the machine excels the human operation. One can also visualize that the machine requires no breaks or time off for lunch and doesn't care whether there is light in the area or not (unless it is equipped with a visual type sensor). It doesn't care whether it is day or night. Thus, perhaps, to use the word emulate is to state the situation somewhat correctly. In this case emulate refers to only those actions that the arm is capable of performing. It obviously cannot duplicate everything a human arm is capable of. This, of course, brings us to the type and kind of task where the human emulates the robot. The robot performs in a delightful manner as long as it is working properly and nothing goes wrong, and as long as all conditions of its task remain the same. But, unless it is an adaptive type robot, any change in the conditions of its job or task will not be accounted for until it has been re-programmed to compensate for those changes. The human, naturally, compensates automatically.

ARTIFICIAL INTELLIGENCE

One definition of artificial is that which is made by man to imitate nature. One can conclude therefore that artificial intelligence is a man-made imitation of nature. One can also assume that an imitation can never be equal to the real thing.

In robotics the actions of a machine may be of such a complex and delicate nature that one almost assumes the machine has a human-like mind direct its actions. Such machines have enough logical capabilities that what they are able to do seems uncanny. But if we fathom the human mind slightly further we find out that there is one aspect of the human mind which, so far as we now know, cannot be duplicated by a machine. That is the ability of the human mind to question. A machine analyzes, performs, has logic, and can make selections, but does it ever ask why? Scientists are learning more about the human brain and its functions all the time. This marvelous electro-chemical device is said to be capable of doing billions of operations per second, of storing incredible amounts of information for extremely long periods of time, and of using such information in the solution of both new and old situations. It has been said that a human is the sum total of what he has learned--and he learns all the time. Since everything is fluid and changing, the human must constantly adapt and learn and adjust to meet changing conditions and situations.

So, some say that the only tasks and operations a human can perform are those that he or she learned to do previously, either by instruction, or by trial and error. Understanding this, one begins to think that a robot might be able to do as much as a human if it has a computer with a large enough memory bank to store all the information that comes to it in the form of learning (structured instruction) or in the form of feedback (trial and error). These trials are generated by some interior command arising from some need which its sensors can determine. The machine then senses a need, and looks through its memory banks for a solution. Finding none, it takes the closest approximation to a solution--some action which will cause the sensor information to change-tries it, and senses the result. If an improvement in the original sensed need results, the direction of the trial is correct. If no improvement results, or the need gets worse, then the direction of the trial is incorrect.

In this example we find the machine doing what many humans might do to solve a problem. But nowhere in this example is the concept of the machine stopping all operations and asking itself "Why is there a need of this kind?" This is what humans will do under similar circumstances. We do not know how to program a machine to ask that kind of question, exactly, although it may be approximated in complex pro grams.

WHO IS PROGRAMMED AND WHY?

Are we all programmed? That is a philosophical question which we, perhaps, might have to consider as we make mechanical machines more nearly like humans. We think of programming as a function related to computers and we think of the microcomputer being closely associated with robotics.

These small, cheap computers have become the foundation behind the concept of robots or automated and intelligent machines. It is said that even computers are a form of robotics. They do have mechanical outputs (printers and graphing devices) as well as visual display outputs and many other output devices. What kinds of orders do they obey? In the beginning such a machine is told to do task one and then do task two and from there it operates on certain conditions. For this example, the orders have been stated as follows.

1. Do f then g

2. If C then f else g

3. While C do f

What is implied, but not apparent, in using these orders is that the computer may be evaluating and scrutinizing masses of data which govern the use of these orders. It is far beyond the ability of the human mind to evaluate this data in the same systematic, logical, unemotional, and timely manner that the computer does in arriving at, say, an implementation of step 2 above. One might suspect that the human mind is programmed to evaluate certain amounts and types of data, and computers are programmed for another type. Human conclusions usually must be a consensus before the conclusions are accepted. We note that machines follow a man specified routine to evaluate even the greatest amounts of data, but can only evaluate those types that man understands how he evaluates! We can think through this problem by imagining that we are in a fast modern aircraft and suddenly, ahead, is another aircraft bearing down on us. Should we dive, climb, turn left, or turn right? What will the other pilot do? What if he maneuvers in such a way that the intercept possibility increases? It is possible that you have had a similar experience while you were walking down the sidewalk. Another human approached. You moved to your left, and he moved at the same time to his right. You moved to your right, and he moved to his left. There was no way you two could prevent bumping into each other. The consequences here were not disastrous; they would be disastrous in an airplane. Could a computer have solved the situation in either case? The FAA says yes.

They have installed collision avoidance computers and control systems (robots, if you will), on board test aircraft for further evaluation. The results look very promising. Collision avoidance systems are used on board ships at the present time.

What can the machine do that the human cannot do under these conditions? It has the ability to evaluate more data, faster, and evaluate it in a more complex routine than the human mind is capable of. For example, using radar sensors the computer can determine in an instant if the other aircraft is going up or down relative to your position. It can determine the speed of movement in all directions and predict or anticipate collision or no collision. The solution to the problem is then flashed to the human in a form he can most quickly and accurately assimilate-perhaps a light to indicate the direction he should steer his aircraft. The job is done well and safely by the two minds, electronic-mechanical and electrochemical, working together! In a case where the human response time is too long, delayed, or cannot be accomplished in the time interval necessary, the secondary part of the machine operation-the robot steering-takes over and for a few precious seconds handles the aircraft. Aircraft autopilots, which steer aircraft along precision courses, at given altitudes, have long been recognized and accepted. They are being given additional duties all the time and tests have been conducted flying a passenger type aircraft from takeoff to touch-down without human hands ever touching the controls.

THE HUMAN MEMORY PROBLEM

The problem is that we forget. We also cannot assimilate much information at one time. It comes into our minds slowly, we retain it when we use it, and we lose if it we neglect using it for any length of time. Computers normally won't do that.

Failure can mean the loss of memory in a computer, but normally the computer can accept, sort, and store all kinds of information quickly and for a very long period of time. Many computer memories today are magnetic tapes, magnetic bubbles, or various types of discs. Some are solid-state units which have information "burned" into them, so that when energized they immediately form voltages and currents which translate into data.

Sixty executives, all presidents of corporations were asked to select a number from zero to nine and they were asked to do this 100 times each. At the same time a computer was programmed to select, at random, a number from zero to nine . , and it was permitted to do this 100 times. It was predicted that the presidents would choose the same number at least 10 percent of the time. That was not exactly correct. They selected the same number 12.3 percent of the time; at least those who had increased their corporate profits during the previous operating years did this. Those presidents whose companies lost money selected the same number only 8.3 percent of the time. The computer selected the same number 10 percent of the time as predicted by laws of probability. This test was said to indicate that the successful executive had strong and accurate intuition and the results were reported in Time magazine. When these results were analyzed it was said that man's inefficiency, emotions, and the nuances with which a man reads data may inhibit his thinking or affect the accuracy of his conclusions. Using a computer to predict trends or to select courses of action based on evaluations of reams of data and then tempering the prediction with human judgment seems to be the best way to gain the utmost from the two worlds.

When we consider the world of advanced robotics, then, we must concern ourselves with the problem of how much intelligence we can build into a machine, how it will use that capability, and when we must insert ourselves into the control loops to achieve the best of the two worlds. We have to decide when the machine can function alone, and when it must be aided by human intelligence. Of course, it goes without saying that we will be forever trying to fabricate a complete mechanical-electronic-chemical mind which will permit humans to specify our commands and relax while everything is done for us! It has been shown by R.S. Aha of Grumman Aerospace that man's thinking process might be specified as shown in Fig. 5. In this block diagram form we can easily imagine how the brain of a very intelligent robot might function.

Fig. 5. One concept of the human thinking process.

MORE ON THE GRIVET ROBOT

We learn still more from a deeper examination of the Grivet robot's characteristics. The arm sections as well as the whole arm itself are moved by direct current stepping motors which have built-in integral shaft encoders. This means that a motor can drive any arm section to within plus or minus .004 inch of a given position when the same command is re-issued. The motors are driven by a single-board microcomputer which can be connected to, and accepts data from, external lines. This computer has a built-in diagnostic capability to pinpoint machine-operational problems. The microcomputer uses a nonvolatile memory cartridge for program storage. Its controller is a 16-line, optically isolated, TTL-compatible unit.

Although the arm is electrically positioned, the gripper, or end-effector or hand, is pneumatically operated. The robot has a built-in air compressor to power this end unit. The use of pneumatics means very fast responses are possible.

ROLE OF MATHEMATICS IN ROBOT-ARM OPERATION

In an earlier discussion we mentioned the algorithm, or equation which a machine's computer solves to permit the robotic machine to function as the designers desired it should function. Considering the Grivet robot arm, we can gain a good insight into just what this means.

In Fig. 3 we examined the teaching pendant as the little box is called. On it we have seen that there are switches to control various movements of the various segments of the arm. We imagine, then, that when we want the arm to do a particular task, we just move the proper button and watch that segment of the arm move until it is in the proper position for the next segment to be moved. We next cause that segment to move until it reaches the proper position, and then we change switches to cause the next segment to move, and so on.

What if we make a mistake? If we are teaching the machine by using the little pendant to control the arm, what happens if we "overshoot" the desired position and have to do some manipulating and jockeying around to get the arm segment into the proper position? Will the robot arm then also jockey around each time it moves? What happens in this case? You can imagine that a person manipulating that arm for the first time would certainly not be able to handle the control pendant switches so perfectly that the arm would move exactly and precisely to each desired and necessary position for the completion of the required task. You merely need to try your hand at radio control of some small model with reasonably tight control to verify this idea.

So, what happens? You guessed it! Within a robot is placed a bit of artificial intelligence that is given the prime directive to assist both the human and the arm so that each, given its own faults, will not inhibit correct operation. The human fault is jockeying the controls when teaching. If the robot has a fault it would be its eagerness to respond, which it does with alacrity. So the computer's task is to eliminate the human fault and compensate for the robot fault to bring about a perfectly delightful operation which is pleasing to the human and maybe to the robot also! Let us look at the computer and bring it under tight scrutiny for a moment. We know its job will be to solve equations and in this case they have to be equations of motion.

We also know that when solving an equation of motion there must always be a reference point. In this case it is the base pedestal of the robot arm. Imagine that this pedestal is set into the floor, or mounted thereto in such a manner that it points north and thus the directions of south, east, and west are also defined. Another way of looking at this coordinate designation would be to say that north equals forward, south becomes backward, or reverse, east is right and west is left of the base structure. We can so define this arrangement if we imagine ourselves being located in the space occupied by the pedestal and have the specified body relationships.

From our location we look at the end point, or gripper, or tool to be positioned which is mechanically fastened to the end of the wrist. That is the point which we want to move to a given location in compliance with a series of commands. We will, of course, also want that end tool or gripper or whatever to do something at each end point. It might pick up an object, it might set down and release an object, it might position a welding element, it might operate a paint sprayer, it might draw a path from the beginning of movement to the end movement such as is done in graphics, it might wire-wrap a terminal or a series of terminals, etc. But the end point must move from somewhere to somewhere else very accurately and in such a manner that a duplication of its action will also be exact-perhaps for thousands of movement operations. Its repeatable positioning must be present and very accurate.

Remember the Grivet operation is within plus or minus .004 inch! So what does the computer do? The computer remembers the starting point and the end point of the gripper for each cycle of operation. A cycle will be defined here as that series of movements required to perform one task. The computer then computes the best path through three-dimensional space for the gripper to move to accomplish the task the human led it through so laboriously, and the resulting path equation or algorithm will be solved by the computer each time it runs the arm through that cycle. We thus learn that the gripper may not go through the same points in three dimensional space that we put it through when we taught it where it was to move. It might move along an entirely different set of spatial coordinates to accomplish the same task better and more efficiently. It definitely will not remember any jockeying we might have done, or manipulating we might have had it do to get it where we humans desired it to be when it finally got there. All those little breaks will have been removed from the motion-path commands and the result, to everyone's delight, will be a smooth, direct operation.

WHAT IS A TASK?

Fig. 6. Vectors describe the position of the end point' position of each arm segment.

We have used the word task and it may have a different meaning to different people. A task may mean one completed action, or one complete cycle. Here it means a movement of an arm-section point to one specified position in a series of end-point positions in order to accomplish a job. The end point is the tip of the gripper or hand. Many tasks make a cycle. We consider one job accomplished when one cycle has been completed.

A vector is a line that has both direction and magnitude.

In Fig. 6 we find three vectors with fixed lengths-V1, V2, and V3 -and these represent the three arm segments of a robot such as the Grivet. They are attached to each other and to the base at one point as shown. Each arm segment can move independently. In this two-dimensional drawing all of the angles cannot be shown. VI moves up or down by opening or closing the angle alpha (a). V2 moves left or right (as shown here) by opening or closing the angle beta (ß). V3 moves up or down, and so moves its tip which we imagine to be the robot's gripper, by opening or closing the angle gamma (y). With a little imagination you can visualize the simultaneous movement of all three vectors such that the end point might be moved up, down, left, right, or in a combination of these directions as required by the tasks. Of course, V3 could be longer so that if the angle gamma were reduced to zero, the end point would come around to the first section point and, equipment permitting, would coincide with that position. As shown, the end point has been moved up 12 units and out along the positive axis 26 units.

We might imagine that the end point is a welding flame or contact point for an electric welding unit. We might also imagine that we want the end point to move up from where it is shown to point above it, and that when doing this movement, the arc will be energized by another command so that welding would take place along the strip T-M. In your mind's eye, visualize how the vectors must move through their angles to keep that end point pressed against the metal being welded along the strip T-M. Angle beta will increase, as will angle gamma. It is possible, but not probable that the angle alpha would increase during this operation.

This is a two-dimensional drawing using just the up down and left-right movements of the end point. In actual practice a robot arm will move in three dimensions, so that there would be a rotation of the vectors into and out from the paper also. This might happen if the end point had a gripper which was to pick up something from a conveyor belt on one side and move it to another conveyor belt located on the other side. The mathematics of the operation would then have to be such that the gripper might move down along the T-M line as shown, until the gripper tip was level with the origin point (0) or in direct contact with the X axis.

In the Grivet-type robot - and others of this type-the arm movement is specified to the computer in the form of the first section point, the second section point, and the third section or gripper point. When these points have been fed into the computer for all the tasks needed to complete the job, and in the proper sequence, the computer will solve the necessary trigonometric relationships for the best movement to reach each end point. If you need to refresh your memory concerning task and job definitions this is a good time to go back and do so.

GIVING THE ROBOT SOME INTELLIGENCE

We have indicated that a robot arm might be able to pick up objects from one conveyor belt and move them to another belt. There might be many objects on the first belt and the arm might be programmed just to pick up selected items and not all of them. There are two basic methods of accomplishing this. First, give the robot some eyes, and then match the shape of things on the belt against memorized shapes in the robot's memory. The memory bank here might be a cassette tape, so that if you wanted to change the shape of objects recognized, you would simply change the cassette software.

Another method might be by identification of the position of the object on the first conveyor belt. With a constant rate of movement by the conveyor, and precision dumping of objects onto its surface, the objects should come by the robot station in precisely separated distances. The arm could then be easily programmed to pick up anything (within its gripper capability) spaced that distance apart, move it, and return for the next object. A problem might arise if the object belt happened to slow down, the objects happened to be placed on the belt incorrectly so the gripper could not easily grasp them, the conveyor belt happened to speed up for some reason.

One use of this idea is the welding of automobile bodies by robots. The bodies come by at precise speeds and precise positions. The arms move to precise positions also and weld and weld and weld. .. and do a good job of it! In robot systems where some intelligence in the form of contact feedback is incorporated, if the body happens to be just a little out of line, the arm will move to compensate for this discrepancy, also a speed monitoring system on the conveyor track or whatever, will keep the robot system informed so that even if the line speeds up or slows down, the arm will adjust to compensate for this change. No doubt you have thought of other ways in which the robot might sense or determine or find objects on that first conveyor belt. Some other means which have been considered are: temperature of the body, size of the body, and-believe it or not-actual recognition of the body even if it happens to be in some unusual orientation on the belt! This latter case is very important because it means that such robots can actually detect and pick out specified objects among many other objects on a big tray or on the belt surface.

This is important in some assembly-type operations.

Giving the robot intelligence then means that it will have a computer to solve motion movements, it will have sensors to assist it in accomplishing its job, and it will have some kind of anticipation and adjustment circuit in case its program doesn't exactly fit every situation.

In many current situations where robot arms, or robots with fixed locations, are used and there are a multitude of robots "employed" on a line, it is possible to have one intelligent robot controlling the operations of many other dumb robots. The tasks may be similar, and the jobs may be the same. In some cases the intelligent robot may control another robot in such a manner that the second robot does some operations or jobs which assist or complete the job the first robot is performing. This is the case when one robot with one arm actually needs two arms to do the job. The solution is to get a second one-armed robot and program it from the first one so that the necessary actions are then accomplished.

Another form of intelligence built into a robot such as the Grivet is that of delay. This means, in human terms, the programming necessary to cause the arm to wait for an object to arrive on the conveyor belt, if it has not arrived when the arm is moving to pick it up. Robot arms can move very fast. It may be necessary to tell the robot to move its arm into a pickup position and then wait a specified time before moving to actually pick up the object. Suppose a robot arm puts on a bottle cap and then has to screw it down on the bottle. That takes time. The arm must not move the bottle until the screwing operation has been completed.

THE ROBOT DIAGNOSTICS

It has been said of an intelligent robot that it, like its human counterpart, is always checking itself to see if it is okay. A human begins to moan when it gets a sore throat, or when muscles ache, or when the stomach is upset. A robot moans also, but in a different tone, as its sound comes from a bell, or siren, or screeching of gears, or clang of parts. Both the machine and the human must have some means of diagnostics to find out what the cause of the problem may be.

With humans there is conversation and a probe into "memory banks" to find related experiences to give clues to the trouble. Second there is a probe of memory banks to find out what to do in case the problem justifies a more serious approach than an aspirin, bed rest, or other nice homey remedies. Probably the diagnostics will start with a tempera ture reading, then pulse reading, then X rays, and so on. Of course it will involve a probe of the memory banks of some good physician who will try to relate the problem to some thing in his experience. He will then probe for the cure used or effective treatment to reduce or eliminate the problem.

With robots the approach is much the same. Within a good robot system there will be sensors to monitor the various moving parts. Simple remedies will be found in the minds of attending humans once the robot has voiced its complaint by means of its sensors and indicators. Within the intelligent robot's brain will be found memory banks that know what the motor speeds should be, when a motor or other shaft has cracked or broken, when there is an oil leak or pressure leak in a flexible line, or that there is no voltage or incorrect voltage to some critical point in the system. The memory banks recognize these problems because they have been programmed with the correct operational conditions, and thus, by comparison they can determine when something is wrong.

Some robot systems constantly monitor all parts of their systems just to make certain that a failure does not occur.

When the machine senses that something is going wrong, it may alert its human operator and say that it needs maintenance, or it may shut itself down while keeping its indicators on to tell its human companion where to look and what to do to get it going again. These kinds of systems run a diagnostic check on themselves when they are first energized to see that everything is all right.

In a previous paragraph we discussed the need for a test system to test the tester, etc. If you combine that knowledge with this series of paragraphs you can determine the difference in requirements. Here the memory banks of the computers have the necessary knowledge to know when an output is correct or not. If the monitoring feedback, for example, does not correlate with what is in the memory banks, either be cause the memory bank information has changed or the monitoring information has exceeded specified limits, then the system shuts down or reduces operation to a safe value.

Meantime there is an automatic alarm to the human super visor to do some checking and find out what the problem might be. Alarms might be visual, in a high sound-density situation, or even be a remote paging system in case the human is not around. With the modern technological advancement in synthetic speech capability, the machine might just start speaking to its human supervisor: "I've got a pain in my tentacle!"

ROBOT COMMUNICATION

Let's expand voice communication slightly and think about robots talking to robots. Some people write about the danger of having robots communicate with each other. One writer says, "Many Americans fear this new age of computers that talk to other computers and operate machines. Technology is moving so fast, in so many areas, that people are afraid of it because they are not familiar with it." In assembly line operations where many robots do the same type of job, it is useful for one machine to direct the work of other machines. The G rivet series, for example is so designed that one might have four slaves and one master or supervisor for these four units. It is said that the operation of many such machines is "orchestrated" so that they can per form process control, handle inventories, handle outputs and inputs to stock-age areas, and control the flow of items in batch lots on various types of moving tracks, belts, or whatever. Ask yourself what that word orchestrated means? If we devote some thought to it, we realize that in a large orchestra there are many different instruments, different parts to the melody and harmonic effects which all blend together in order to produce the final, ear pleasing sound. Thus we arrive at the conclusion that orchestrating a group of robots means the blending together of many different processes, automatically performed in a timely, coordinated manner to bring about the fabrication of something, or the completion of some process comprised of many parts. It could, of course, also mean the systematic and coordinated demolition of something if that were the process desired! Robots take many forms even when they work singly as shown in Fig. 7.

Yes, robots on assembly lines talk to one another. They may communicate that they have completed a task, they may ask for help from a nearby machine so that two arms can be brought to bear on a process instead of the one with which each robot is normally equipped. They may govern the flow of materials or regulate the inspection and processing of the materials as required. We can imagine one robot at the end of an assembly line, trying to fit some parts together and finding it difficult, communicating via its own computer and connecting lines back to the robot which is responsible for machining the parts: "Hey Buster! Get on the ball down there! Reduce your product by 1/10,000,000 inch!" We can also imagine an old robot instructing a new one in the tasks it must perform.

One big problem in manufacturing is that there are many small and different batches of parts and items which must be made. This means, of course, that the robot assembly line must change its actions and operations whenever a new batch is to be handled. Consider an assembly line of up to 30 robots in length. What a way to get into a new dimensional sys tem-30 robots in length! Normally, people would simply use new approaches and methods to handle the new and different batches as they come down the line. What do the robots do? If there is a change in products and there is good and orchestrated communication between and among the robots concerned, a human merely inserts a new program cartridge into the master computer. The human changes the memory for just one robot in this system and does not have to change the memory of all the robots as he or she might have to do if they were operating as individual units. It is advisable to have good, complete communication among the robots in manufacturing or processing plants! People who fear these connections, will learn that the communication is not intelligent reasoning, as sometimes takes place when humans communicate. Among robots, the communication consists of a series of drive and feedback signals, which cause the robot receiving the information to do something. When it does what it is supposed to do the reverse communication takes place which informs the master that it did what it was supposed to do, and to what degree of precision it did it. If the reverse communication doesn't take place, the master knows the slave isn't working properly.

COSTS

Of course robots cost money. They start at $10,000.00 per unit for manufacturing types, and $3,000 for some of the android types we will discuss in another section in this guide.

Android types are hobbyist's toys right now, but imagine that someday they will be household servants and let us all live life in Utopia! But, what about costs in the manufacturing situation? One might evaluate the cost of programming and operating a robot vs the cost of using human hands to do the job in cost per hour. Remember that the robot doesn't complain, doesn't worry, doesn't take rest breaks or lunch breaks, is always on time, and can work 24 hours a day without getting tired (worn-out maybe, but not tired). Each robot has an operating cost per hour which can be compared to the human cost per hour and when the purchase has been amortized the relative costs of the two will show something interesting. Although the robot won't ask for a raise, its operating costs will in crease somewhat simply because the costs of energy will increase, and it uses energy. It doesn't get tired, but it should have some kind of normal maintenance routine and humans do not normally have to be maintained on the job. It is true that the robot machines will do lots of jobs humans don't like to do and it will do them well. This will release humans to do other, more profitable work. So, if humans are to become future Robot Masters then they must prepare themselves now for profitable and rewarding jobs. Technology will not stand still.

Fig. 1.7

Fig. 8. Feedback accuracy and measurement.

FEEDBACK ACCURACY

We show, in Fig. 8, a kind of illustration which may indicate how the accuracy of measurement of the end device position may affect the general operation of the robot as a whole.

Along the X axis we have the physical position which can be measured by the feedback potentiometer, ac wave com parison unit, pressure feedback, or whatever. Along the Y axis we have the accuracy of repeatable operations which are performed time after time after time. Note that we never reach a perfect operational state. If nothing else, Murphy's Laws will prevent that from happening. But we do reach a close proximity to the state of blissful perfection, providing that the feedback measuring elements each can provide the high degree of accuracy required.

We have shown a linear drop off of accuracy. This may not be exactly correct. Each machine may have a different curve even though each machine may be constructed exactly the same. The operational environment, the type of load, the use to which the robot machine is put may well affect its repeatable accuracy. From this illustration, then, we begin to get some understanding of the complexity needed and the precision required in order to exactly locate and assemble small batch parts, or to perform operations where a few millimeters of movement may be the whole ball park. We are reminded that difficult measurements are not exactly impossible. Back in the Dark Ages of the 50's when the atomic bomb was being developed, there came a requirement to measure extremely small amounts of U-235. Using a delicate, and almost unthinkably precise, balance the scientists concerned were able to measure U-235 in amounts of millionths of an ounce. Of course you would not even dream that such a measuring unit might be used in other than a laboratory operation-or would you? Think what that kind of measurement possibility would mean in terms of precision feedback of an advanced robot! Now we are imagining the use of a machine robot to do jobs that are being performed by humans using microscopes.

This requires them to move with slow, delicate precision because the slightest error could be costly or irreversible.

We imagine that if such precision could be accomplished by robots the costs of many items of modern technology might be reduced and their reliability might be vastly increased. The key to this precision is the ability of the machine to function on the smallest of signals, and to physically measure the smallest value we can think of.

This leads to another problem area-maybe it's Murphy's Laws again. The smaller the signal or the more sensitive the feedback, the greater the danger from extraneous electrical and mechanical noise. Noise means error signals and noise is the enemy which must be overcome in the fight for greater precision and better computation.

SPEED OF OPERATION

In a study of servomechanisms, which form the basis for all robotic operations, we find that the greater the speed of movement the harder it is to prevent overshoot and oscillation. There are, of course, mathematical solutions to many of these kinds of problem, and they are used. We want to plan beyond the present day capability and consider what the truly maximum speed of operation of a robot system or an individual robot might be. We are considering power behind the movement, and such amounts of power that the overshoot and oscillation problems again rear their ugly heads.

How fast is fast? Why do we need to increase the speed of operation? Is it useful, or required, or just a desirable development? Is productivity, and therefore profits, dependent upon speed of operation? Do we want to work toward that ultimate blinding speed of assembly, or are we going as fast as we can go considering our knowledge, and capability of making advanced robot machines. Consider the CYRO 5 + 2 advanced robot made by Advanced Robotics Corporation.

This machine has two coordinated arms instead of one to accomplish welding operations. One hand has a five-axis torch motion (x,y,z,c,a,) and the second hand has two-axis standard motion. Both hands are said to move smoothly and accurately to accomplish many tasks. Of course the coordinated motions must be computer directed or program directed.

Computers are being improved every day and their speed of operation, output capability, and controllability in crease every minute. It seems that we must expect physical motions from our Advanced Robot machines to keep pace, and these motions, at vastly increased speeds, must be accomplished without overshoot or oscillations. Also required will be the ability to perform, for long periods of time, these fast and precise operations, i.e. a high degree of reliability and very low down time ratios. In Fig. 9 we illustrate the GYRO 5 + 2 type robot and some motions and jobs associated with it.

It is interesting that this robot is said to move in rectilinear coordinates. One must envision that this is accomplished through computerization of the angular type coordinates and movements which evolve from the movement of the end device shown at B. It is said that rectilinear type construction provides a stiffer structure for greater accuracy and smoothness than would a comparable angular movement type. Also, it is claimed that a rectilinear system provides greater safety since people tend to think in rectilinear coordinates and thus can anticipate the machines motions.

Fig. 9. (A) Cyro 5-plus-2 two arm advanced robot. (B) Movement of "end unit" of Cryo 5-plus-2 (courtesy Advanced Robotics Corp.).

Fig. 10

Fig. 11

If one examines Fig. 9 (A) it is easy to see what is meant by the rectilinear motion. The arm moves up and down, in and out, and its support column can move left and right.

There is no turning of the arm or its support column elements.

There are rotary motions associated with the end gripper or, in this case, the welder unit. No doubt some will say that there are certain advantages for the rotary movement systems. That, of course, will be a function of what kind of machine is being manufactured and its particular application. would suspect that there is a use for each type of machine, and that usage may govern what type advanced robot is procured by a manufacturer.

SAFETY

In advanced robotics we must always be conscious of safety just as we would be conscious of this requirement in any plant or operation where machines are used. For the advanced type robots we have so far considered, we recall that they are usually fixed in place, that they have a certain radius of operation, or space volume in which their actions are confined, and thus, regardless of how they move or what they do, if we restrict ourselves from entering that action space around each robot, we should be safe enough.

If there is a malfunction which should, by some very unique and very unusual condition, cause the separation of the physical parts of a robot from its base structure, then, of course, that might make unsafe any other volume of space around the unit. But this is a case which is so rare that one might consider it just doesn't happen. Normal safety procedures such as are well developed and practiced in machine containing plants and operations, will normally suffice, and will protect those humans involved when they are in the same physical areas as the advanced type robotic machines. One would wear protective glasses if the machine is producing particles of any type which might be injurious, or if the rays from the operation (welding) might be injurious to the eyes.

Clothing must be such that sparks or whatever type of remnants might be cast off from the tasks performed will not ignite or penetrate or cause other troubles to the human contained therein. Proper footwear would be very nice to have over your toes just in case you are close and the gripper opens prematurely, dropping a large object to the floor. Normally, however, one would be walking or spending much time near the robot assemblies. Observation is through instruments and TV type tubes and visual inspection from isolated and protected control booths. Only in the case of trouble with a particular machine is human presence required in its particular space. Then adequate precautions are taken when initiating the repairs and that will serve until that day when the advanced type robots repair themselves! The mobile type robot may present another situation. Its volume of active space may also be confined but that might be a large space and a moving space and one must then practice safety as one does in the street with moving automobiles! We can just imagine our reaction if we are in a large building and suddenly, silently, this monster machine comes bearing down on us with its four arms waving madly. The arms are moving--not because it is angry, for robots don't get that way- because that is the way advanced robots normally move! Of course we could be on the regular robot track as defined by some invisible substance on the floor and it may be just moving from one job location to another in accord with some of its prime directives. Yes, unless it is programmed to stop for obstructions it might well run over us! But even that, somehow seems remote as a possibility. Certainly, any advanced robot of the mobile class will have such sensors that it will stop and inquire-perhaps verbally-if it encounters anything unusual in its path, or on its job, or during its performance of its normally scheduled and programmed tasks.

Fig. 12. Voice synthesizer chip (courtesy Texas Instruments).

ROBOTIC SPEECH

Not many years ago it was considered almost impossible to make a machine which could actually talk back to its operator. As we well know, that impossibility has been re placed by fact. One of the companies in the forefront of speech synthesis in Texas Instruments, whose Speak and Spell solid state learning aid has started the machine talking revolution.

Figure 1-12 shows the chip which resulted from many years of research and development in the speech producing effort.

The heart of TI's Solid State Speech systems is this little monolithic speech synthesis chip which was invented by TI. It actually generates electronic signals which, when reproduced on a loudspeaker system, sound like the human voice. The chip - models the characteristics of the human vocal tract.

Machines that speak, either to acknowledge or convey information, are deemed the next area of the technological revolution according to TI. They are engaging in a big pro gram of research and development related to computers, industrial machines, telecommunications, automobiles, and to the entertainment market. Now even an automobile, properly equipped with a small diagnostic computer, will tell the mechanic what is wrong or what needs adjustment in or on an engine. How nice that concept is for the backyard mechanics! If the machines will only explain carefully and in detail exactly how to correct the problem, as well as telling you what the problem is, then we've got it made! That you will find more and more speaking machines in every phase of our lifestyle is almost guaranteed by such indicators as TI's separate speech organization to serve as a focal point for all their new applications and developments in this field. Let's take a closer look at that chip by examining Fig. 13. You can amaze yourself by examining Fig. 12 and locating the Pipeline Multiplier chip and then looking at Fig. 13 to see what is inside that little device! But as you see from Fig. 13, that's only a part of the circuitry involved.

Fig. 13

TI defines synthetic speech as either a word, phrase, or sentence, or a complete or unique sound. A lot depends on its use, duration, and application. A word, for example, is defined as a second of "utterances", and only TI can explain what is meant by that! Perhaps that is related to the 100 word vocabulary of the chip in Fig. 12, which can be spoken or uttered in 100 seconds. Of course improvements in vocabulary, and if required, speed of making the sounds, will be produced as the need arises. We are truly approaching that day when all commands and instructions to our advanced robots will be given by simple speech commands, and the robots will also advise us of their condition and when they need maintenance they will provide us with any other information pertinent to the overall job, simply by telling us verbally, or vocally, or making some sounds! - As pointed out by TI it is relatively easy to increase a robot's or machine's vocabulary. This can be done in many instances simply by appending a sound to an existing sound.

In the polled-status mode, the host CPU (Central Processing Unit) issues the address of a word, sets the talk command, and polls this status bit to determine that the word has been spoken. Delays between words and sentences are inserted by addressing the particular delay word which is processed as if it was just another word.

The monolithic speech synthesis chip uses Linear Predictive Coding (LPC), which duplicates the human vocal tract.

As this name implies, LPC is a linear equation which formulates a mathematical model of the human vocal tract. Thus it is possible to predict a speech sample based on previous samples. LPC is a technique of analyzing and synthesizing human speech by determining from the original speech a description of the time varying pitch and energy using digital filters which also reproduce human sounds when excited by random or periodic inputs. Because digital impulses themselves cannot be used to drive loudspeakers, it is necessary to have, on that same chip, a Digital-to-Analog (D/A) converter which transforms the digital information into signals required to energize the loudspeaker or earphone. On the chip shown in Fig. 12, an 8-bit Digital-to-Analog Converter is used. This is identified in the lower right hand corner of Fig. 13. Codes for the twelve synthesis parameters (10 filter coefficients, 1 pitch, and 1 energy) serve as inputs to the synthesizer chip. These codes may be stored in a ROM (Read Only Memory). When the codes are decoded by on-chip circuitry (Fig. 13) they produce the time varying signals descriptive of the LPC model, or human voice sounds.

The input to the digital filter takes two forms, as we have stated. They may be periodic or random. The periodic input is used to reproduce voiced sounds that have a definite pitch such as vowels or voiced fricatives (sounds formed and pronounced by forcing the breath through a narrow opening between the teeth, lips, etc, such as f, s, v, and z). In the TI the fricatives are z, b, or d. A random input to the chip models unvoiced sounds such as s, f, t and sh. The speech synthesis chip has two separate logic blocks which generate the voiced and unvoiced sound patterns. The output from the digital filter, drives the Digital-to-Analog Converter and that drives an amplifier which drives the speaker. The rest of the chip, as shown in Fig. 13, consists of the integrated array multiplier the advanced 10-stage lattice filter, the adder and multiplier, and the delay circuits. A complex bit of circuitry, but since it is all on a chip it is easy to obtain and use.

A final note about the speed of operation. With the unit shown (and newer units will have a higher speed of operation) the rate is about two inputs per each five microseconds. TI says it takes twenty multiply and accumulate operations to generate each speech sample, but the circuits can still generate up to 10,000 speech samples per second!

SUMMARY

Throughout this section I have presented a glimpse of what the world of robotics consists of. We have found that it is a vast world indeed, encompassing mechanical engineering, electronic-electrical engineering, light, sound, chemistry, atomic and computer engineering. Some disciplines which we have not mentioned may be found in this field. So we learn that it is a big area of study, design, and development. We have called advanced robotics a science and an art and we have presented a simple type of definition: "The design, use, and operation of machines, which are computer controlled to do human desired tasks." We also indicated that within the control framework might be a human brain as a part of the computing system.

There is an organization, dedicated to manufacturing type robotics which has given a somewhat different type definition of a robot. The Robot Institute of America has come up with a typical engineering type definition: A reprogrammable, multifunction manipulator designed to move material, parts, tools, or specialized devices, through various programmed motions to perform a variety of tasks. The manipulator in this case is what we have called the end product (or tool) holder, or gripper.

While it is true that robots of any type, industrial or not, will get smarter as time progresses and will be able, through use of many sensors which humans do not have, to accomplish many tasks relating to jobs we have for them, in a most pleasing manner, they still must be programmed by someone.

The user may not develop the software necessary, but some one has to develop it, and someone has to develop the command instructions so that an android robot, for example, will respond to your spoken commands.

Webster's dictionary has another definition of a robot, and from this can be derived another definition of the world of robotics: "A robot is a machine in the form of a human being that performs the mechanical functions of a human being." Of course, this has been the general concept carried about in many human minds for a long time, but it is not necessarily a true definition as we apply it to today's electromechanical marvels. The Madison Avenue personnel devote hours and hours to research and planning and burn much midnight oil to prepare sales pitches which involve the magic word "robot(s)", and in many instances what they define as a robot, for sales purposes, is not a robot at all. We have to make allowances for the kind of sales personnel who jump on the bandwagon, and present things which are not exactly correct, according to the definitions presented herein.

It is very interesting that the Lord Company of Erie, Pennsylvania hopes to market, a robot hand made out of a kind of sponge and filled with a grid of wires and sensors which are so arranged that it will have an almost human sense of touch and feel! Robots are here now! Let's move ahead in our study of some advanced types, and some hobby types.