A Swiss team has applied Darwinian selection to robot development, producing robots that can walk, cooperate and even hunt each other.
"Just a few hundred generations of selection are sufficient to allow robots to evolve collision-free movement, homing, sophisticated predator versus prey strategies, coadaptation of brains and bodies, cooperation, and even altruism," say the Ecole Polytechnique Fédérale de Lausanne and University of Lausanne researchers.
"In all cases this occurred via selection in robots controlled by a simple neural network, which mutated randomly."
The input neurons of the neural network were activated by the robot's sensors, and the output neurons controlled its motors.
Each robot had a different 'genome', describing different connections between neurons. This resulted in unique behaviour and fitness - how fast and straight it moved, for example, or how often it collided with obstacles.
At the beginning, the robots had random values for their genes, leading to completely random behaviours.
But Darwinian selection was then imitated, by choosing the genomes of the robots with the highest fitness to produce the next generation.
To do this, genomes were paired, and random mutations such as character substitution, insertion, deletion, or duplication were applied.
The team found that within 100 generations, robots were able to move without collisions in a maze.
When 'breeding' for predator behaviour, a range of strategies evolved, including lying in wait and circling the walls.
And, astonishingly, the robots were even able to evolve altruistic behaviour, in a task that involved pushing tokens around. Some could be pushed single-handed, earning the robot one 'fitness point'; others required two robots, gaining the whole group one point.
It was found that groups of unrelated robots - those with randomly differing genomes - invariably took the selfish approach and went for the small tokens. But those with similar genomes generally pushed the larger tokens, cooperating to raise the fitness of the whole group - and thus reducing their own chances of 'winning'.
"These examples of experimental evolution with robots verify the power of evolution by mutation, recombination, and natural selection," conclude the authors.
"The ability of robots to orientate, escape predators, and even cooperate is particularly remarkable given that they had deliberately simple genotypes directly mapped into the connection weights of neural networks comprising only a few dozen neurons."
The full report is in PLoS, here.