Bush Robots Hans P. Moravec Carnegie-Mellon University July 30, 1980 @heading(Clumsy Claws) Present day remote controlled and robot arms fall far short of human dexterity. Besides basic limitations in power/mass ratio, they lack sufficient degrees of freedom, especially in their "hands", to manipulate complicated objects, or apply the combinations of forces needed to accomplish most simple tasks. These limitations (and others) exclude them from many interesting areas. Analysing and imitating the human hand is one approach towards lifting these constraints. A hand with three dexterous fingers able to roll an egg and twirl a baton (slowly) has been demonstrated by the Electrotechnical Laboratory in Japan. Presently such manipulators lack the precision being achieved by more conventional robot designs. There are tasks for which even human hands are inadequate. If these are attempted at all, it is with varying degrees of success using special tools and fixtures. @heading(Fine Fingers) It may be possible in future to leapfrog the dexterity not only of conventional mechanical manipulators, but of human hands as well. Consider the following observation. Once upon a time animals were shaped like sticks (worms), and couldn't manipulate or even locomote very well. Then the sticks grew smaller sticks and locomotion was much improved, and manipulation a little. Then the smaller sticks grew yet smaller sticks, and hands were invented, and manipulation got better. Generalize the concept. I visualize a robot that looks like a tree, with a big stem, repeatedly branching into thinner, shorter and more numerous twigs, finally ending up in jillions of near-microscopic cilia. Each intermediate branch would have three or four degrees of freedom, an azimuth-elevation mount at its base, and an axial rotation joint at the top, where it connects to the next level of smaller twigs, and possibly also a length altering telescoping joint. To a large extent fewer degrees of freedom per level can be traded off for more levels. Each branch would also incorporate force sensing. Though each branch would be a rigid "mechanical" object, the overall structure would have an "organic" flexibility because of the great multitude of degrees of freedom. For concreteness consider a particular design; a twenty level tree where each branch connects to two, each half the size. There would be a million tiny end effectors, each one millionth the size of the trunk, in linear dimension. Under most reasonable assumptions about the nature of the "muscles" of the tree, the smallest branches can oscillate a million times as fast as the trunk. The tree is made of only twenty different sizes of branch, ideal for mass production. I imagine that a finished robot would be able to construct all but the smallest of its own component branches. The smallest twigs may have to be made by some sort of three dimensional lithographic technique. Since volume increases with the cube of dimension, each level masses only one fourth as much as the next larger level, and all the cilia together mass only one trillionth the mass of the trunk. The touch sensing data rate into such a device is comparable to the data rate of human vision. Even more interesting, the mechanical effector data rate out of the robot is also on the order of billions of bits per second. Compare this to the human hand which puts data into the world at perhaps a few hundred bits per second. Such high data rates imply huge amounts of co-ordinating processing power (again comparable to human vision), but imagine the possibilities. The tree robot could reach into a complicated piece of delicate mechanical equipment, simultaneously sense the relative position of a host of parts, and rearrange them for a near instantaneous repair. In most cases the super touch sense would totally substitute for vision, and the super dexterity would substitute for special tools. It is almost certain that the branches must contain some local processing power. If they also contained energy storage and means for remote communication, the tree could be mobile, walking on its branches, and smaller robots could be obtained by detaching subtrees. It could even walk on ceilings with the tiny cilia holding onto cracks. Subtree detaching and branch replacement would also be the self-repair process. Here are some more "practical" thoughts. I imagine power and data flowing from level to level. The data moves serially on coax, perhaps. Thicker coax can handle higher frequencies, thus the lower levels of the tree can handle the data concentration. Same for power. If the robot is mobile it needs internal power sources and control. Each segment can have a battery and a computer, but both may be too little to provide all the services that segment needs. If so, it makes sense to tap power and control from the much bigger segments one or two levels down towards the stem of the bush. The battery and the computer two levels down are 64 times as massive, providing 64/4=16 times as much energy and crunch power for the 4 segments at the upper levels. If you use this strategy, the first level or two can't run themselves. They can provide a stable framework if they passively lock their joints. @heading(So What?) It is pretty clear that a full fledged bush robot can't be built in the near future, so why worry about it? One answer is that if progress in mechanical construction were to catch up with the rate of improvement of electronics, namely capabilities doubled each year, then one could imagine adding one level of branches to such a robot annually. In perhaps fifteen years we would have a twenty leveller. Emerging, bootstrapping, robotics and application of microelectronic construction techniques to mechanics may make this fantasy come true soon. Even if this is not the case, the control problems of such a spectacularly redundant mechanical system have a fascination in a more abstract sense. I would like to investigate co-ordination strategies for bush robots at CMU, and possibly even build a simple three or four level working model. My current idea is for a recursive problem solver which passes plausible incomplete plans, with constraints, to right and left subtrees (things like, Left tree: stay on left side of plane A, and apply net force vector V to object. Right tree: stay on right side of plane A and apply force -V to object). If a subtree problem can't be solved, a complaint is passed back to originating node, which tries something else. The theoretical work will be a lot of fun because the robot can be simulated on the 3D modelling machine which is the subject of another proposal. This multiprocessor can detect collisions of the modelled objects, and also produce realistic shaded graphic representations of them. Imagine a movie of a realistic looking 15 level bush picking up and moving objects, walking, or doing hula dances. @heading(Fabulous Fiction) Bob Forward is using the Bush Robot idea for his second science fiction novel. Here are some of his observations. @i[ I recently sat down and documented the ideas for the mobile "repair" robot that you suggested for the "hands" for the computers that I use in Roche World. This is a section out of the "hearings" appendix that is the science fact background for the novel. Still a first draft, so needs work. I expect that the "shoulder" imps will become an important part of each of the human characters behavior. (Why pause to scratch your ear when you have someone that can do it for you?) @heading(@i(The Christmas Bush)) The hands and eyes of the near-human computers that ran the various vehicles on the expedition were embodied in a mobile extension of the computer, popularly called the "Christmas Bush" because of the twinkling lights on the bushy multi-branched structure. The development of the bushlike shape for the repair and maintenance robot has a parallel in the development of life forms on earth. Once upon a time, the most complex animal was a worm. The stick-like shape was poorly adapted for manipulation and even locomotion. Then these stick-like animals grew smaller sticks, called legs, and locomotion was much improved, although they were still poor at manipulating. Then the smaller sticks grew yet smaller sticks, and hands, with manipulating fingers were invented and precise manipulation of the environment became possible. The Christmas Bush is a manifold extension of this concept. The motile has a main trunk that repeatedly branches into thinner, shorter, and more numerous twigs, finally ending up with millions of near-microscopic cilia. Each intermediate branch has four to six degrees of freedom, and the force-sensing needed to handle large rugged objects, like door handles and pieces of equipment, as well as the delicate insides of a precision instrument. Each segment has some small amount of intelligence, but is mostly motor and communication system. The segments communicate with each other through their interconnections to coordinate their activities, but the main computer in the vehicle is the primary controller, communicating with the various portions of the bush through coded laser beams. It is the colored lasers sparkling from the various branches of the bush that give it the appearance of a Christmas tree. It takes a great deal of computational power to operate the many limbs of the bush, but the built-in "reflex" intelligence in the various levels of segmentation lessen the load on the main computer somewhat. The analogy of the isolated brain with its massive intelligence communicating with the finger muscles on the hand out at the end of the arm through nerve impulses is a good one, where the laser beams act the role of the nerve impulses. The Christmas Bush, however, has capabilities that go way beyond that of the human hand. The bush can stick a "hand" inside a delicate piece of equipment, and using its lasers and detectors as light source and "eyes" rearrange the parts inside by feel and "sight" for a near instantaneous repair. The bush also has the ability to detach portions of itself to make smaller motiles. These could walk around on their larger appendages, and even walk up the walls and along the ceilings with the tiny cilia holding onto microscopic cracks in the surface. The smaller twigs on the bush are capable of very rapid motion. In free fall, these rapidly beating twigs allow the bush to propel itself through the air. Even at earth gravity, the smaller sub-trees can fly using their cilia. The speed of motion of the cilia is high enough that the tree can generate sound and can talk directly with the humans. Another interesting property of the bush is its ability to change size. Just as a human can go from a crouch to an erect position with the hands and fingers raised and change its height from less than one meter to almost three meters, the bush can shrink or stretch by almost a factor of five, from a short, squat bush to a tall, slender tree. Each human has a small sub-tree that stays with the human to act as the communication link to the main computer. Most of the crew have the tiny bushlet ride on their shoulder, although some of the women prefer to keep them in their hairdo. In addition to acting as the communication link to the computer, the imps also act as health monitors and personal servants. Their assistance is especially helpful in keeping the crewmembers hair in place when in a spacesuit, and are the ideal solution to the perennial problem of spacesuits, scratching an itchy nose. Yes, the imps go into the spacesuit with the humans, and more than one human life was saved on the mission by an imp detecting and repairing a suit failure or patching a leak. In fact, there are two computer motiles with each suited human. The personal one that stays with the human inside the suit, and the suit motile, usually larger in size, that stays with the suit. It is usually outside in the life-support backpack, but can worm its way inside through the air supply hose. We think that life would be weird with a semi-living creature always attached to us, yet think how bereft you feel when you have forgotten your eyeglasses, or pen, or wristwatch, or computer. I am tempted to make the computer and its appendages the "viewer" of the story. But since it cannot be too intelligent (in that case, why take humans along?), it might be a little hard for the average reader to "identify" with. Bob]