You might not toss your kids on the ash heap but a “mother robot” trained to evaluate the progress of her child bots will do just that.
Roboticists have developed a “mother” robot that can build and evaluate her own “children,” and then decide which version performs best to inform the design of the next generation. Remarkably, the system doesn’t require any human intervention.
Here’s how the system works: A so-called “mother” robot is programmed to build a “child” robot that’s capable of rudimentary locomotion. This child can consist of anywhere from one to five plastic cubes, each with a small motor inside. Then, without any human intervention or computer simulation, the mother robot evaluates the quality of her offspring according to a speed test, and then uses that information to inform the design of next generation of progeny. It’s survival of the fittest, but applied to robots.
“Natural selection is basically reproduction, assessment, reproduction, assessment and so on,” noted lead researcher Fumiya Iida in a statement. “That’s essentially what this robot is doing—we can actually watch the improvement and diversification of the species.”
In five separate experiments, the mother designed, constructed, and evaluated ten agents over ten generations (for a total of 100 candidates). Each experiment typically began with a randomly generated child-bot. As the experiment progressed, the mother robot mutated her offspring by manipulating the physical configurations of the five blocks, which in this case can be construed as the robotic equivalent of genes.
By the time the mother robot got to the last generation, her spawn performed a speed task twice as quickly as the best individuals in the first generation. What’s more, her ability to improve performance increased over time. The researchers say this was on account of the robot’s ability to fine-tune design parameters during later generations.
Fascinatingly, the researchers say some designs weren’t likely to have been conceived by a human; it was truly doing it’s own thing.
The science is complex. You can read the original study at Morphological Evolution of Physical Robots through Model-Free Phenotype Development.
Mike “Mish” Shedlock