Self-development and resulting improvement of the species have long been at the core of human activities. Interestingly, such traits are somewhat familiar in all areas of animal kingdom. One of the greatest attributes of intelligent animals is that of “adaptation”. We, humans can adapt, and so can many other species. Imagine some robots adapting to the environment or job requirements.
Let’s imagine a situation, you bought a dwarf but fat domestic robot (like R2-D2) to clean surfaces and walls of your home. It does that job perfectly. One evening you return home and find your robot has reassembled itself and has grown few inches taller and became something like a humanoid droid (like C-3PO droid). All because, it could not reach the ceiling of the room that needed cleaning. Daydreaming, you say?
Artificial Physical Evolution
Researchers at laboratories at ETH Zurich and University of Cambridge has published some results which promises something just like that. They had built and experimented with a robotic schema, which can perform “autonomous” construction of new robots through “trial and error”. This robot schema, which researchers are calling “mother robot”, can produce multiple generations of “child robots”. The mother-bot keeps the best performing “child-bots” and rebuilds or “improves” inferior builds. I think this whole process is somewhat similar to what some animals follow.
The optimization is based on an evolutionary algorithm, which iteratively improves the construction sequence of locomotion agents. An ordered set of one to five genes contains the complete information describing an individual, the genome. The algorithm starts initializing the genomes of the first generation of candidate agents. Evolution then is performed through mutation, where components in one gene are modified or single genes are added or deleted, and crossover, where a new genome is formed by merging genes from two individuals. The genome then is interpreted and expressed into the phenotype which describes the appearance of the final agent.(Brodbeck et al. 2015)
The authors used a phenotype-gene-genome relationship in achieving improved generations of robots. The mother robot measured how far can a child travel before it runs out of power. In a chart published the authors have shown how the fitness have varied among different “species” or “varieties”.
What future holds?
Although the test cases used for the study only used very small scale robot, the concept and algorithm surely promises to mimic the biological concepts of evolution and speciation. The idea itself is thrilling when considered in context of recent space missions and the question of whether or not to send manned missions to far ends of our solar system.
With enough mechanic materials in stock our space stations can become “robot colonies” instead of “colonies of human scientists”. Whilst this has enormous potential of coming in the aid of human species, at the same time, should we be worried as a species? Erroneous human species have done more damages to humanity and the globe, could accurate robot mummies with their own “will” be a threat to us? Perhaps not now, but in the words of scaremongers, we better watch our skills. 🙂