The German high tech research Fraunhofer-Institut has just unveiled its Genetic Robots. The robots are made by using genetic algorithms that come up with the optimal robot shape and form without the involvement of any human designer. The optimal form is decided based on a physics engine that takes into account the tasks and terrain and then designs the robot accordingly. The robots are made of ball and socket joints and connecting tubes and can change shape depending on the required tasks. The resulting robot is then 3D printed.
This is nothing short of the future of design. By looking at the fitness of a design before the robot is made and by individually designing and 3D printing only the best suited designs, more optimal designs can be produced. No longer do designs have to be “for all terrains” or for all uses but they can suffice for a single task. In effect you have disposable highly specialized robots. With Moore’s law and other technology accelerators in effect this commodity approach has long seemed to me to be more fruitful in the long run than making some kind of humanoid “mega Asimo” robot that can do everything.
Removing the human designer from the equation has profound implications for design. What will human designers do? Is this just a fad or a single application or will human designers be pushed back from more and more products as genetic algorithms combined with physics engines take over? Will this kind of evidence based automatic design be able to make beautiful things? Will it be able to, through number crunching, come to design solutions that work in many areas of our world? And how well will designs that have been simulated beforehand work in the real world?
I think that this is an astounding project by the Fraunhofer-Institut that will really show us the way forward in design, manufacturing and 3D printing. One of the largest issues with 3D printing is the lack of design skills and lack of inspiration that many people experience. They want to make something that is perfect for them but do not know how. Will machines fill the gap and design for us? Also, John Connor, are you out there?
CORRECTION: as pointed out on Boing Boing, I omitted to mention that the Fraunhofer Institut”s work is based on research conducted by H. Lipson and J. B. Pollack and their GOLLEM project. You can read about that here.
The robot in that picture looks like Tetra, from Brandeis””s ten year old Golem project. New Scientist reported on that project in 2000, what””s the deal?
http://www.demo.cs.brandeis.edu/golem/
Yes, I””ve seen it now also. I””m a huge fan of Hod and his work but didn””t know about GOLEM previously. They do mention the GOLEM project on the Fraunhofer site though. So perhaps their innovation lies in the execution part of making the robot?
I””m really excited that somebody is picking up where the Brandeis researchers left off. What a truly COOL project!
This series does not “remove the designer,” per se–it just redefines the role of the designer. Rather than providing a solution, the designer is tasked with defining the appropriate problem to solve. A genetic algorithm uses a fitness criteria to evaluate different prospective designs; the final design is only as good as the fitness criteria allows. Some fitness criteria are easy to define: no hard edges, center of gravity, minimum shear strength. Others, such as aesthetic quality, are harder to define in an equation.
A genetic algorithm excels at finding surprising solutions, nonintuitive answers to difficult problems. But there is no universal fitness criteria. Every single problem requires a human touch to evaluate and retune the algorithm to help it get in the right direction. Genetic algorithms are merely a tool, albeit a quite interesting and surprising tool.
Genetic algorithms do not technically require tuning. And you can just use a self-tuning GA anyway. And if that is too expensive for you, then just remember that mutation should be low (50%) and selection should be stronger the higher the mutation rate is.
Most of your time should be spent making sure that it is easy to define new parts so that robot designs can be updated as new materials and parts become available.
Love it!
Looks similar to something I worked on, but I didn””t have 3D printers back then: http://studie.erikdebruijn.nl/thesis/ (scroll down all the way for Evomorph. In the paper there are more details about the artificial neural networks and morphology that are evolved genetically.
The convergence between the virtual and the real. It means that, increasingly, dreams do come true.
Erik,
I””m waiting for people to be able to imagine a thing that can then become a design and later be produced automatically.
Turning an idea into a product effortlessly.
Needless to say, what a terrific website and informative posts, I””ll add backlink – bookmark this internet site? Regards,
Reader.
While very well.I protest against it.Your phrase is matchless…For the life of me, I do not know.Should you tell.
Wow! Thank you! I permanently needed to write on my blog something like that. Can I implement a portion of your post to my site?