Mother robot programmed to make smarter robot children independently of humans

Evolutionary robotics have made another step towards fully autonomous machines.
A team of researchers from the University of Cambridge and ETH Zürich has developed a ‘mother robot’ which is capable of producing new robots from available components, test their performance, learn from its mistakes and improve the next generation of its’ babies’ without any human intervention. This might give us a hint about the evolution of robots in the future.

Scientists – Luzius Broadbeck, Simon Hauser and Fumiya Iida – have published the results of their experiments in a journal PLOS One. They have programmed a ‘mother robot’, which looks like a robotic arm, to build a series of simple robots ( 500 robots over ten generations)and optimize their performance by manipulating their configuration in order to get them to cross a pre-determined distance as fast as possible. The whole process of designing, building and testing each individual robot lasted for about 10 minutes.

The ‘mother robot’ was able to analyze data collected through the observation of its ‘children’ in a way that no human can, and generate a new concept of the next generation, by preserving those traits that have been proven good in the previous generation and modifying those which had negative impact on their performance, thus simulating natural process of evolution on these artificial beings.

Having in mind that the whole experiment was performed with some rudimentary components – plastic cubes with a built-in motor inside, the results this autonomous robotic arm achieved were more than surprising. As it moved through the experiments, abilities of mother robot were gradually increasing and in the final stage the whole generation of its creations performed better than the best ones from the first generation.

This “model-free technology of artificial evolution” uses no computer simulation or any other specific type of instructions given to the robot about how to perform its task, but rather ‘motivates’ them to produce better robots by a programmed ’reward’ function.
Combined with a sophisticated AI, and a wider range of assembly parts, these types of machines will be able to do wonders in the future.