Researcher Atilim Gunes Baydin has published a paper describing the evolution of central pattern
generators for the control of a five-link planar bipedal walking
mechanism (PDF format). A physical walking platform was built that
had four joints - a knee and hip joint on each leg. Initial attempts
were made to hand-tune a stable walking gait to see if it was possible.
This proved very difficult with the walker destabilizing after a short
time. Next, controllers were evolved using genetic algorithms in a
simulated environment and then tested on the real hardware to see how
they performed in the real world. This produced a number of stable,
human-like gaits including some which were able to maintain a stable
gait even when not making use of sensory feedback.
First of all, bravo to those who made the robot and software.
Really, though genetic algorithms have been around for some time to do just this very thing, to help home in on a great design for gaits and navigation, so I'm not sure what the fanfare is here? This would be a classic way of attacking this sort of problem. It's great that they had a purpose, that they programmed that purpose and design into the software tool and it did what they asked. It's a good thing that they were able to program in a purpose and a design they wanted or else without those things the algorithm would have never been able to come up with something useful. This is a great example of the designing process and finding the best tool to help with that design.
The design vs evolution thing, though, always seems that they hype the evolution slant and try to push the design part under the rug. It's as if when you use a genetic algorithm you aren't doing any designing and poof a cool thing just emerges mysteriously somehow. If you have a purpose to create the best gait design, you create a tool, program and plug in the parameters that you want, and lo and behold it does what you asked it to do, you get results. How is that really proving evolution and not intelligent design? What is the threshold? When does a purposeful design all of a sudden becomes evolution? If no initial design parameters were plugged in and no purpose were given to the tool then what would the algorithm do? Nothing at all. They wanted a tool to do something and it did what they wanted. The task was too time consuming, tedius or too monumental to design such a walking gate by hand so they let the computer work on homing in on the best design. This is vastly different than evolution! If they gave it direction of what to design even generally speaking, and it did it, there's not much to get excited about in terms of evolution. It is impossible to come up with a great design without some sort of directional purpose. When you remove the designer and or purpose then the algorithm can do nothing if it doesn't know what to design or has no purpose. The genetic algorithm is a design tool and it helped to design within the parameters given.
There again, I see the merits of using such a genetic algorithm to speed up the design process intelligently. I mean ultimately, it's a design tool. I guess I just have hang ups on the terminology, calling it evolution when it's a purpose driven, directed process of elimination algorithm. For many people the word "design" is taboo and the word "evolution" is the paramount, when for me it's obviously just the opposite. "Intelligent Design" gets no respect. Genetic algorithms and evolution are all the rave, when really they are nothing without a design and a purpose.