These robots can breed with Darwin in mind

The design improved generation after generation as the computer learnt to discard features which were useless or cumbersome — just like the process of natural selection that prevails among life forms in nature. After several dozen generations, the smartest robots in the simulation were able to make clumsy movements. After several hundred more generations, the computer selected the three `fittest’ designs and then built them in real life, using a 3D Printer — a nozzle that builds up progressive layers of heated thermoplastic that cools down and solidifies, forming a three-dimensional structure. The only help from humans was to insert a small standard electrical motor enabling the robots to move.

The machines are small, simple and far removed from the popular vision of a robot: they are made of simple triangles and rectangles, like a child’s construction set, and move forward by pushing a bar or limb against the floor. Even so, they are a major achievement as they cross the robotic threshold from virtual into physical reality.

“This is, to our knowledge, the first time any artificial evolution system has been connected to an automatic physical construction system,” say the inventors, describing their brainchild as “a primitive replicating robot”.

The work by Hod Lipson and Jordan Pollack appears in tomorrow’s issue of Nature, the British scientific weekly. In an adjoining commentary, robotics pioneer Rodney Brooks acclaims the work as “a long awaited and necessary step towards the ultimate dream of self-evolving machines”.

Artificial intelligence is a long-standing dream of scientists, although it also has a controversial side, evoking dark images of machines which could eventually gain supremacy over man. Even after several decades of improvement, robots remain quite dumb. Human help is still needed to build and maintain them, and the cost has to be recovered through mass production, such as in toys and manufacturing systems.

A robot that could learn from its environment but also melt down its own parts and change its shape would be an ideal candidate for exploring other planets, where terrain, atmosphere and radiation could be hostile.

In another article in Nature, researchers from the University of Lausanne in Switzerland announce that they have taught robots community spirit, training them to behave like ants. Small robots were programmed with some of the simple rules that ants use when foraging for food: when to leave the `nest’, how to avoid bumping into each other, how to search for food and how to tell their comrades about rich finds of resources. This work, too, has an intriguing futuristic role. Swarms of robots, with autonomous, decentralised organisation and cooperative abilities, could be used to explore other planets which are too distant or hostile for humans.