Starfish creatures shuffle in long lines towards the concrete walls of the deserted nuclear power plant that have been overgrown with climbing plants and weeds. Slowly waddling on their long, soft legs, they disappear through the hole in the wall towards the reactor. The creatures that enter the dangerous space undisturbed are robots. But unlike the robots that already populate the world, they are not designed by people. These originated from evolution.
Mother Nature has proven over the past four billion years that evolution offers a solution for every design problem: by constantly adapting creations, letting go of them in the world and seeing who survives, minute changes ultimately lead to bodies that can be found in every conceivable situation. See man. The squid. The bat.
Could not we use the principle of evolution to make robots that are much more focused on their task than a human designer can think of? The human designer is still trapped in the frame of thought where we quickly reach out on arms and legs, or – do wildly – on a large dam disc on wheels that sucks the floor. Leave the design process to Mother Nature and who knows what kinds of robots you do not get.
Guszti Eiben, professor of artificial intelligence at the VU University in Amsterdam, has not let go of this idea for seven years. What if Mother Nature is replaced by a supercomputer that rapidly develops virtual robots, releases them in a virtual environment, where they have to perform virtual tasks and then the best inventions ‘survive’ and are allowed to ‘reproduce’ with other survivors?
Nice idea. It already exists. But Eiben wants to develop it a step further. What if you let materials materialize in the real world, to see how they behave? Build the limbs that the computer has created using a 3D printer, connect them to motors and sensors and release them.
What Eiben wants has never been done before. Everyone he told about his idea loved it. But no research financier wanted to support the plan. NWO rejected Eibens research proposals. The professor bought an inexpensive 3D printer and built out of a primordial soup of servo motors and circuit boards a small robot world with robots that partly came from the simple 3D printer. Two years ago he showed the world the first baby robot and a working proof of concept .
Again: everyone enthusiastic. And again it became quiet. Especially because Eiben actually wanted more. Suppose, he says, that you have really good 3D printers. And supercomputers. Then you can release the entire design process. You can develop robot bodies in every conceivable form. Focused on every conceivable task. There will be robotic beings that exceed our wildest fantasies. To be able to perform those tasks that we now consider impossible. Agriculture on Mars. The mining of gold on the seabed. Or the dismantling of abandoned nuclear power plants.
The latter are in full force. Britain alone has eleven power stations that have been put out of operation and are waiting for decommissioning. Some of the tribes from the fifties. The designers have died and there are no more construction drawings of some. Their interior is a nuclear maze with probably collapsed parts, puddles of oil and water and, who knows, deep cracks in the concrete floors. See to that safe once disassemble.
According to an estimate by the British Nuclear Decommissioning Authority, the decommissioning of nineteen existing nuclear facilities will require 100 billion pounds over the next 100 years. A study by the University of Birmingham even last year amounted to 300 billion euros. The process is costly, it takes a lot of time and exactly how it should be done, nobody knows.
Eiben, with his view on robot evolution now a good acquaintance in the scientific robot world thanks to many publications in Nature , among other things , could have some of the solution with his idea. What if we let robots evolve until their bodies and properties are optimal for performing clean-up tasks in the abandoned complexes?
Conditions for robots are that they are small enough to enter through narrow openings, and flexible enough so that they can find their way between the probable junk, trenches, caverns, potentially dangerous liquids and possibly collapsed parts. Should they be able to walk, drive or roll? Should they be able to climb, grab and dig? Should they have hard claws, or soft, octopus-like tentacles? Again: no one who knows.
And that is not necessary either. If the development of the cleaners is left to Mother Nature or its digital counterpart, it will be fine. Or according to Eiben’s motto: evolution is the best designer.
Together with British colleagues from the universities of Bristol, Edinburgh and York, the VU professor wrote a research proposal for the Engineering and Physical Sciences Research Council (EPSRC), the British counterpart of the Dutch research financier NWO. Each university has its own specialty: Bristol has two hundred robot scientists and industrial 3D printers. York has materials scientists who are going to work on new plastics with which the hard parts of the robots (the mechatronics) can be developed. In Edinburgh there is a supercomputer that can create a parallel world in which robots and their evolution are simulated.
After the plan was submitted, another period of waiting started. Almost unexpectedly, on a Friday afternoon in December, an email came from Great Britain. It seemed that the EPSRC Eibens wanted to finance plans. There was money, a lot of money: converted more than 2 million euros, to be divided among the three British research institutes, each of which will carry out their own part. Exactly according to the vision that Eiben already formulated years ago. The professor made an air jump in his office and organized a drink for close employees that same afternoon. His robot evolution could begin.
Now he is still beaming in his office, still a bit worried. The research, which starts in August, will arise from three parts, Eiben sketches. There will be a physical environment that consists of a birth clinic and test rooms. There robots will be fabricated with a 3D printer and separate parts such as sensors, motors and chips. These mechatronics are the organs of the robots, and are picked from an organ bank. How the robots will look like: nobody who knows.
The ‘robot babies’ will first be trained in a school, just like people, so that they can acquire basic skills, such as walking. The successful types are released after a learning period in a ‘nuclear cave’ without radiation, with pipes and rubbish on the ground, where the robots have to find their way, observed by the researchers. Their task: mapping the environment and cleaning up things. “The devices will not grow bigger than a football,” says Eiben. This has a practical reason: they are easier to build and they will soon have to go through small openings in a real reactor.
Simultaneously there will be a virtual space that is an exact copy of the real ‘cave’, with the same kind of virtual robots, whose behavior is simulated by a supercomputer. This computer will create new variants of the robots according to evolutionary models and release them in the virtual space.
The great thing about this design is that we can cross successful real robots with successful virtual variants’, says Eiben. If virtual robot children in the virtual world are successful, the researchers can decide to have such a robot child ‘born in real life’, from the 3D printer, so that they can scurry around in the real world. In this way the advantage of computer evolution (fast and many) is linked to practical evolution (slower and more realistic). ‘By combining both we can reduce the so-called reality gap.’
Virtual robots can be crossed with real robots, real with real and virtual with virtual. ‘The genetic code in the real world must be identical to that in the virtual. This allows us to cross both, ‘says Eiben. ‘This was one of my deeper insights when I worked on the concept.’ In this evolutionary pressure cooker, evolution can get a boost and in a relatively short period of time usable robots can be developed, is the idea.
Evolution is certainly not the best designer in all cases, says Andy Tyrrell, professor of electronic engineering at the University of York and co-researcher. That’s right. Evolution has never conceived a hyena on wheels, while in some circumstances the wheel offers great advantages.
‘But’, continues Tyrrell, ‘evolutionary processes are particularly good at creating effective designs when little is known about the system in which they have to operate, and where possible conflicting goals exist for which there is no single solution.’ Exactly the environment in which these robots have to work.
According to Eiben, the beauty of the project lies in the two phases it consists of: the first part consists of engineering, the second of which is science. An ideal combination, says Eiben. ‘First you will develop the instrument, to do experiments with it afterwards.’ The instrument, the so-called evosphere, where robots evolve, learn and survive in order to receive offspring after all, does not yet exist. Eiben: ‘Just as an astronomer needs a telescope to study the stars, and a nuclear physicist a cyclotron, a developer of evolutionary robots needs an evosphere.’
The evosphere, with its 3D printers and supercomputers, will be built in the first half of the four-year research project, in the second period the robot evolution will take place in practice. The ultimate goal: autonomous colonies of self-propagating robots, preferably in a closed loop in which worn or unusable robots are reused without human supervision. Whether it comes to a working concept is uncertain. ‘The two most difficult areas are the integration between hardware and software and the development of an automated assembly system’, says Tyrrell.
The idea of robots that procreate itself causes fear to some people. “There are people who think it is dangerous what we do,” says Alan Winfield, professor of robotics at the University of Bristol and one of the project researchers. ‘But what we do is actually artificial selection; what farmers have been doing for millennia with crops and livestock. But with robots. ” Moreover, the lab is a closed system, says Winfield. The robots do not just walk into the outside world. ‘We are already lucky if they go anywhere anyway’, he says on the phone. In addition, the robots will get their energy from batteries, and they will be empty soon.
Ook Alex Ellery, universitair docent aan de Canadese Carleton Universiteit en gespecialiseerd in de risico’s van robots die zichzelf kunnen vermenigvuldigen, denkt dat eventuele gevaren van het project voldoende zijn ondervangen. ‘Vanwege de beperkingen van de onderzoeksomgeving en de menselijke greep op de beschikbaarheid van bouwstenen.’ Met andere woorden: de robots zullen niet uit het lab kunnen ontsnappen en als de toevoer van grondstoffen wordt gestaakt, stopt hun evolutie vanzelf. Dat maakt de kans op robots die de wereld overnemen zo goed als nul.
Will robot babies be born anywhere in the world in four years to be used for all kinds of applications? The researchers deem that unlikely. “In four years’ time, we hope to have shown that artificial evolution is possible, and that we have developed equipment and tools that allow us to further develop our ideas,” says Tyrrell.
‘In my wildest dream we discover new aspects in evolution’, Winfield sketches. ‘That we will have more insight into evolution in four years’ time. That would be great. “
Eiben is also optimistic: ‘I hope that we will soon be able to answer existing and new scientific questions.’
On 27 May, Guszti Eiben will give a Paradiso lecture about artificial intelligence, artificial life and robot evolution: Being a robot – that is not a life! (Only in Dutch)