Designing stable and robust controllers for multilegged robots is a
challenging task, and it would therefore be desirable to develop
automated methods for doing it. Learning the control behavior is
difficult however because optimal behavior is not known, and the
search space is too large for reinforcement learning and for
straightforward evolution. As a potential solution, this project
uses an approach called Evolution of Network Symmetry and mOdularity
(ENSO). ENSO generates appropriate gaits for a given robot and
environment by utilizing a mathematical model of animal locomotion
and group-theoretic analysis of controller symmetry. It represents
the controllers as interconnected neural network modules and
optimizes module functionality and interconnection symmetry using
evolution. In this project, this approach is evaluated by evolving
controllers for a quadruped robot in physically realistic
simulations. On flat ground, the resulting controllers are as fast
as those having hand-designed symmetry, and significantly faster
than those without symmetry. On inclined ground, where the
appropriate symmetries are difficult to determine manually, ENSO
produced significantly faster gaits that also generalize better than
those of other approaches. On robots with a more complicated
structure including knee joints, ENSO resulted in more regular gaits
than the other approaches. These results suggest that ENSO can be used
to design effective locomotion controllers for multilegged robots.
Videos of evolved walking behaviors