Evolution of Adaptive Behaviour in Robots by Means of Darwinian Selection

In spite of the apparent behavioral complexity of the robots, all behaviors were achieved with extremely simple brains consisting of only a handful of neurons.



An essay by my two former PhD supervisors summarizing more than a decade of research in Evolutionary Robotics has created quite a buzz in the blogo-sphere:

Ever since Cicero’s De Natura Deorum ii.34., humans have been intrigued by the origin and mechanisms underlying complexity in nature. Darwin suggested that adaptation and complexity could evolve by natural selection acting successively on numerous small, heritable modifications. But is this enough? Here, we describe selected studies of experimental evolution with robots to illustrate how the process of natural selection can lead to the evolution of complex traits such as adaptive behaviours. 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. In all cases this occurred via selection in robots controlled by a simple neural network, which mutated randomly. ...

Once one looks beyond the blogosphere claptrap caused by any mention of words like "evolution" or "predator" in the context of robotics, there are some interesting insights from this body of work: Foremost, the experiments outlined in the essay offer an intriguing real-world illustration of evolution at work. Another is that they offer a powerful way to study biological phenomena such as the evolution of group behaviors like communication or altruism in a highly-controlled, real-world system. Yet another is that embodiment does matter  - not only can the use of robots result in stronger testing of hypotheses and in higher predictive power than purely computational models, in some cases it is the best way to gain insights into the complex behavior resulting from a robot's - or animal's - reaction to the environmental changes caused by its own actions.

However, for all the work's merits I think that its limitations may be even more revealing. In spite of the apparent behavioral complexity of the robots, all behaviors were achieved with extremely simple brains consisting of only a handful of neurons. This is surprising, given that even the brains of animals as simple as the nematode worm C. elegans exceed this by an order of magnitude and animals like the fruit fly Drosophila melanogaster by a factor 10'000. Clearly, we are far from evolving intelligent robots. Furthermore, the performance of robots invariably stagnated after a few hundreds of generations of evolution across all experiments. For now it is unclear if this is a direct cause of using extremely simple brains, or is tied to deeper reasons such as an inadequate genetic encoding. Either way, we are far from (re-?)creating open-ended evolution in the lab.

If you find this interesting, I suggest you have a look at the essay and accompanying videos


Robotics News

Biweekly newsletter on advances and news in robotics, automation, control systems, interviews with leading roboticists, and more.

About the Automaton blog

IEEE Spectrum’s award-winning robotics blog, featuring news, articles, and videos on robots, humanoids, automation, artificial intelligence, and more.