Yasuo Kuniyoshi and Shinsuke Suzuki found that they could get a simulated robot to walk, and even to scamper over obstacles, by programming each leg to obey a chaotic feedback controller.
New Scientist article
They didn’t have to do any conventional programming, or even genetic algorithms, in order to get the thing to walk; they just had to find a good set of parameters for the chaotic controller (also, they added a weight on one side of the robot, to break circular symmetry; the robot was a circle with legs evenly spaced around it).
The article is a little short on details; I am unclear how much care went into choosing a chaotic controller function that would be good for walking. On the one hand it would be more surprising if any old chaotic function seemed to work; on the other hand it might be more useful if there were an aspect of “programmability”.
A brief search found this citation, but I couldn’t find the IROS proceedings online:
Yasuo Kuniyoshi and Shinsuke Suzuki, Dynamic Emergence and Adaptation of Behavior Through Embodiment as Coupled Chaotic Field, Proc. IEEE Int. Conf. on Intelligent Robots and Systems, pp.2042-2049, 2004