Randomly evolved robots

From: mortongr@flash.net
Date: Sun Sep 24 2000 - 13:13:16 EDT

  • Next message: Vandergraaf, Chuck: "RE: What happens when the oil runs out."

    Given all this talk about how randomness can't produce anything good, I thought I would alert everyone to an important development in the creation of evolving robots. Yes this is still early in the process of developing them but an important step has been taken--the creation of robots evolved in a computer and then built by machines--i.e.robots. The article can be found in Aug. 31 Naature. The commentary on the article notes how widespread the use of genetic algorithms are as well as their success. The article says:

    "Over the past 20 years, computer algorithms inspired by genetics2 (genetic algorithms) have become common tools in solving optimization problems. Usually the optimization target is a mathematical procedure of known form, but with many undetermined parameters. A population of candidate procedures is generated by expressing different sets of these parameters as binary strings. The performance of each procedure is evaluated and those that are 'fitter' or do better according to some specific criteria are allowed to reproduce. Reproduction may be sexual, by taking the strings of two fit parent procedures and generating the binary string for the descendant by using cross-over and perhaps mutation, or it may be asexual, in which case just mutation is used. Over many generations, the population tends towards more successful procedures.
    Over the past decade many people have experimented with evolving populations of 'artificial creatures' in simulated environments or virtual worlds. Here, rather than the binary string representing the parameters of some procedure, the string is an artificial genome, which encodes a control circuit (or nervous system) for a simulated robot. So, over time, better-performing robots slowly emerge. In some cases there is an implicit fitness evaluation as creatures fight it out for virtual resources necessary for survival. In other cases there is an explicit 'fitness function' applied to each generation of creatures, forcing evolution in a desired direction.
    Rodney Brooks, “From Robot Dreams to Reality,” Nature 406(Aug 31, 2000):945-947

    And the question of who is a better designer, humans with brains, or genetic algorithms has been decided in favor of evolution at least in the case below:

    “A couple of years ago Paolo Funes and Jordan Pollack5 tried to convert computer models into physical reality by evolving not creatures, just structures — the simulated structures were selected for their strength. They then built physical versions of the structures by hand from real Lego blocks and confirmed that the structures were much stronger than human-designed ones.” Rodney Brooks, “From Robot Dreams to Reality,” Nature 406(Aug 31, 2000):945-947, p. 947

    This analogy to how living systems are created is evident from the following description of the manufacturing technique:

    “First, these particular robots cannot be built by conventional manufacturing techniques. The rapid-prototyping technology solidifies polymers in place, so that ball and socket joints are constructed with the ball already inside the socket. The parts are never separate, and if they were they could not be assembled without damaging them. This is not far removed from the way that biological systems grow. Second, the evolutionary strategy used in these experiments starts with a blank or 'null' genome and randomly mutates it into one that generates a working machine — so there is no in-built bias from seed machines in the population. One could say these machines have evolved 'naturally', without human intervention.” Rodney Brooks, “From Robot Dreams to Reality,” Nature 406(Aug 31, 2000):945-947, p. 947

    At this point I want to note that the genomes of the evolved robots started empty! This should cause pause for those who insist that without the initial input of information to the genome nothing could happen. Mutations placed information into the genomes and the resulting structures were judged solely on how well they moved. The genome decided which structural pieces would be connected and in what way. Lipson and Pollack write:

    “Starting with a population of 200 machines that were composed initially of zero bars and zero neurons, we conducted evolution in simulation. The fitness of a machine was determined by its locomotion ability—the net distance that its centre of mass moved on an infinite plane in a fixed duration. The process iteratively selected fitter machines, created offspring by adding, modifying and removing building blocks, and replaced them into the population (see Methods). This process typically continued for 300 to 600 generations. Both body (morphology) and brain (control) were thus co-evolved simultaneously. “ Hod Lipson and Jordan B. Pollack, “Automatic Design and Manufacture of Robotic Lifeforms,” Nature, 406:(2000):974-978, P.975
    “Selected (virtual) robots out of those with winning performance were then automatically converted into physical objects: their bodies, represented only as points and lines, were first expanded into solid models with ball joints and accommodations for linear motors according to the evolved designs (Fig. 4a). This 'solidifying' stage was performed by an automatic program that combined pre-designed components describing a generic bar, ball joint, and actuator. The virtual solid bodies were then 'materialized' using commercial rapid prototyping technology (Fig. 4b). This machine used a temperature-controlled head to extrude thermoplastic material layer by layer, so that the arbitrarily evolved morphology emerged as a solid three-dimensional structure without tooling or human intervention. The entire pre-assembled machine was fabricated as a single unit, with fine plastic supports connecting between moving parts (Fig. 4c); these supports broke away at first motion. The resulting structures contained complex join
    ts that would be difficult to design or manufacture using traditional methods (Figs 4d and 5). Standard stepper motors were then snapped in, and the evolved neural network was executed on a microcontroller to activate the motors. The physical machines (three to date) then faithfully reproduced their virtual ancestors' behaviour in reality (see Table 1).” Hod Lipson and Jordan B. Pollack, “Automatic Design and Manufacture of Robotic Lifeforms,” Nature, 406:(2000):974-978, P.975-976

    The authors conclude:

    “Without reference to specific organic chemistry, life is an autonomous design process that is in control of a complex set of chemical 'factories', allowing the generation and testing of physical entities that exploit the properties of the medium of their own construction. Using a different medium, namely off-the-shelf rapid manufacturing, and evolutionary design in simulation, we have made progress towards replicating this autonomy of design and manufacture. This is, to our knowledge, the first time any artificial evolution system has been connected to an automatic physical construction system. Taken together, our evolutionary design system, 'solidification' process, and rapid prototyping machine form a primitive replicating robot. Although there are many, many further steps before this technology could become dangerous, we believe that if indeed artificial systems are to ultimately interact and integrate with reality, they cannot remain virtual; it is crucial that they cross the simulation–reality gap to l
    earn, to evolve, and to affect the physical world directly18. Eventually, the evolutionary process must accept feedback from the live performance of its products.” Hod Lipson and Jordan B. Pollack, “Automatic Design and Manufacture of Robotic Lifeforms,” Nature, 406:(2000):974-978, P.977

    All of this was carried out via random mutation, which according to the standard view of anti-evolutionists, shouldn't be able to improve things. Why then do we see improvement when the human isn't drawing the blueprints?

    I would point out that genetic algorithms have been used to create radios from separated parts assembled randomly, structurally stronger landing gear and many other things. Just to head off one possible objection, if anyone wants to cite Dembski's paper from The Nature of Nature conference, they should see what I say on my web page about his paper. Even John Baumgardner, a committed young-earth creationist was appalled at the things Dembski was erroneously saying.

    This archive was generated by hypermail 2b29 : Sun Sep 24 2000 - 13:13:19 EDT