Chapter 12 3.2 Fast, cheap, out of control

The slogan "Fast, Cheap, Out of Control" first appeared on the badges of engineers at conventions and was later used by Rodney Brooks in the title of his sensational paper.The new logic brings a completely different perspective on the machine.There is no control center in a mobile robot swarm.They are scattered in time and space, just like a people across history and continents.Build these droids in large numbers and don't take them too seriously. Rodney Brooks grew up in Australia.Like the other boys, he loved reading science fiction and making toy robots.He had developed the habit of seeing things the other way around, of always acting contrary to conventional wisdom.He kept going in and out of the top robotics research and development laboratories in the United States, pursuing whimsical ideas about robots, and finally accepted the tenured position of the head of the MIT Mobile Robotics Research Project.

There, Brooks embarked on an ambitious graduate project to develop robots that were closer to insects than to dinosaurs.The first to be born was "Allen".His mind is kept on a nearby desktop computer, because that's what robot developers were doing at the time to get a brain worth saving.Allen's body has vision, hearing and touch, and the signals it perceives are transmitted to the "box" containing the brain through several cables.There would be too much electrical background interference on these cables, and Brooks and his team were plagued with frustration.To fix the problem, Brooks switched student after student.They searched every known transmission medium, and even tried various alternatives such as amateur radio, police walkie-talkies, cell phones, etc., but no matter what the solution was, they could not establish a connection that was not disturbed by static interference and could transmit rich and diverse signals. .In the end, Brooks and the students vowed that no matter how small the brain had to be, the next project would be to integrate the brain hub into the robot—so that there would be no need for those troublesome cables.

So when making the last two robots, Tom and Jerry, they were forced to use only very simple logical steps and short and simple connections.Surprisingly, this humble built-in neural circuit actually outperformed the brain when it came to simple tasks.This modest gain prompts Brooks to revisit the foundling "Allen."He later recalled, "As it turned out, Allen's mind really wasn't working." This downsizing gave Brooks a taste of the sweetness and spurred him to continue exploring how dumb robots can go and still do useful work.Eventually, he got a reflex-based intelligence.Robots with this intelligence are no smarter than ants, but they are just as inspiring as ants.

Brooks' vision took shape on a contraption called "Genghis."Genghis is the size of a football and resembles a cockroach.Brooks took his philosophy of parsimony to the extreme.Little Genghis had 6 legs but didn't have anything that could be called a "brain".All 12 motors and 21 sensors are distributed on a decouplable network without a central processor.Yet the interaction between these 12 muscle-like motors and 21 sensors yields an astonishingly complex and life-like behavior. Each of Genghis's tiny legs was working on its own, independent of the rest.Each leg controls its movement through its own set of neurons -- a microprocessor.Each leg just needs to take care of itself!For Genghis, walking was a team project involving at least six little minds at work.The rest of its smaller brain power is responsible for the communication between the legs.Entomologists say that's the solution for ants and cockroaches -- the crawling insects have neurons in their legs that do the thinking for that leg.

In the robotic cockroach Genghis, walking is accomplished through the collective behavior of 12 motors.The two motors on each leg go up and down depending on what the surrounding legs are doing.If they lifted and lowered in the correct order—well, go!One, two, one, one, two, one! ——Just "walk up". No part of this delicate contraption is responsible for walking.Without resorting to a high-level central controller, control gradually converges from the bottom layer.Brooks calls it "bottom-up control."Bottom-up walking, bottom-up alert.If one of the cockroach's limbs is broken, it will immediately adjust its gait and crawl with the remaining five limbs without disorderly stepping.Such transformations are not amputated and relearned; they are instant self-reorganization.If you cripple one of Genghis's legs, the remaining five legs that can still walk will reorganize and walk, just like a cockroach, easily finding a new gait.

In one of his treatises, Brooks first described how to make creatures walk "unconsciously": There is no so-called central controller to instruct the body where to place the foot, or how high to lift the leg when stepping over an obstacle.In fact, each leg is entitled to some simple movements, and each leg can independently judge how to behave in different situations.For example, the awareness of one basic movement is, "If I am the leg and I am going up, then I am going to fall", while the awareness of another basic movement can be described as, "If I am the leg moving forward, I will go down." Let those five guys hold off a little bit."These consciousnesses exist independently and stand by at any time, triggering once the prerequisites for perception are met.Next, to walk, simply lift the legs in sequence (this is the only place where central control is likely to be required).As soon as one leg is lifted, it automatically swings forward and then falls.And the forward swinging motion triggers the remaining leg to move back slightly.Since those legs are just touching the ground, the body moves forward.

Once the robot is able to walk steadily on a smooth surface, you can add some other animations to make it walk better.For Genghis to climb over the phone book lying across the floor, a pair of tentacles would need to be installed to relay information from the ground back to the first set of legs.Signals from the whiskers can inhibit the movement of the motors.The rule might be, "If you feel something, I stop; otherwise I keep going." While Genghis learned to climb over obstacles, his basic walking patterns were not disturbed in the slightest.In doing so, Brooks illustrated a universal biological principle—a divine law: When a system works, don't mess with it; build on it.In natural systems, improvement is "patching" on an existing well-tuned system.The original layer continues to function without even noticing (or having to notice) that there is a new layer above it.

When your friend tells you which way to take to get to his house, he never tells you "don't crash your car" along the way, even if you do have to follow that admonition.They don't need to communicate with you about that low-level goal, because your driving skills already guarantee that goal will be easily achieved.Which way to go to his house is a high-level activity. Animals (during evolution) learn in a similar way.The same goes for Brooks' mobile robot.They learn to navigate a complex world by building a hierarchy of behaviors, roughly in the following order: avoid touching objects

wandering aimlessly explore the world construct inner map Pay attention to environmental changes planning travel plans Anticipate changes and revise plans accordingly The department responsible for aimless roaming will not make a fuss when encountering an obstacle, because the department responsible for avoiding touching objects has already dealt with it. Graduate students at the Brooks Mobile Robotics Lab built a waste-picking robot they cheerfully dubbed the "Scavenger Hobby Machine"—it roams the lab at night collecting empty soda cans.Its aimless roaming part lets it dangle from room to room; the avoidance part keeps it from bumping into furniture as it roams.

Gathering Hobbies hangs around all night until its camera detects an object in the shape of a soda can on a table.The signal triggers the wheels of the mobile robot, propelling it right in front of the beverage can.The arm of the collecting hobby machine does not need to wait for instructions from the central brain (it has no brain) to "understand" its location through the surrounding environment.It has wires attached to its arms that carry signals so that the arms can "see" the wheels.If it senses, "Hey, my wheels stopped," it knows, "There must be a soda can in front of me."So it stretched out its arm to get the jar.If the jar is heavier than the empty jar, leave it on the table; if it is as light as the empty jar, take it away.The robot continues to roam aimlessly with the empty can in hand (it doesn't bump into walls or furniture because of the help of avoiding the department), until it stumbles upon a recycling bin.At this time, the wheels stopped in front of the recycling bin.The silly arm will "check" if its hand is holding a can, and if it is, it will throw it into the recycling bin.If not, just roam around the office again until you find the next jar.

This ridiculous, "hit and miss" recycling system is horribly inefficient.But night after night, with nothing else to do, this goofy but reliable scavenger managed to collect a decent amount of aluminum cans. If some new behaviors are added to the original working collector machine, more complex systems can be developed.This is how complexity accumulates by superimposing rather than changing its basic structure.The bottom-level behavior is not disturbed.Once the aimless roaming module is tuned and working well, it should never be changed.Even if this aimless roaming module prevents new advanced behaviors, the rules it applies will only be suppressed, not deleted.The code is forever, it's just ignored.What a bureaucratic yet biological approach! Furthermore, each part of the system (department, staff, rules, behavior) is functioning without error - as if it were a separate system. The “avoid-touch department” works on its own, regardless of whether the “jar-taking department” is working or not. The "Take the Jar Department" does its job, whether the "Avoid Touch Department" is doing it or not.Even if the frog's head falls off, its legs will still twitch, that's why. The distributed control structure Brooks devised for robots came to be known as the "containment architecture," because higher-level behavior needs to contain lower-level behavior when it wants to dominate. If you think of the country as a machine, you can build it like this with an inclusive architecture: You start with the township.Tackle the township's logistics first: the basics include repairing streets, laying water and electricity pipes, providing lighting, and enacting laws.Once you have some functioning towns, you can create counties.On the basis of ensuring the normal operation of the township, you set up courts, prisons and schools within the county, adding a layer of complexity on top of the township level.Even if the institutions of the county and county disappear, it will not affect the normal operation of the township.When the number of counties is large, the state level can be added.The states are responsible for collecting taxes, while allowing counties to continue to exercise most of their powers.Townships could survive without states, though perhaps not as efficiently or as sophisticatedly.When the number of states increases, the federal government can be added.The federal level subsumes some of the activities of the states by placing limits on the activities of the states and hosting the organizational work above that level.Even if the federal government disappears, hundreds of towns and cities will continue to do their own local work—repairing streets, laying water and electricity pipes, providing lighting.But when rural jobs are subsumed by the states, and eventually by the federation, these rural jobs show even greater efficacy.The towns organized by this inclusive structure can not only build buildings, but also set up education systems, set rules, and be more prosperous than before.The federal structure of the US government is an inclusive structure.