Using evolutionary artificial neural networks to design hierarchical animat nervous systems.

摘要

The research presented in this thesis examines the area of control systems for
\nrobots or animats (animal-like robots). Existing systems have problems in that they
\nrequire a great deal of manual design or are limited to performing jobs of a single
\ntype. For these reasons, a better solution is desired.
\nThe system studied here is an Artificial Nervous System (ANS) which is
\nbiologically inspired; it is arranged as a hierarchy of layers containing modules
\noperating in parallel. The ANS model has been developed to be flexible, scalable,
\nextensible and modular. The ANS can be implemented using any suitable
\ntechnology, for many different environments.
\nThe implementation focused on the two lowest layers (the reflex and action
\nlayers) of the ANS, which are concerned with control and rhythmic movement.
\nBoth layers were realised as Artificial Neural Networks (ANN) which were created
\nusing Evolutionary Algorithms (EAs). The task of the reflex layer was to control the
\nposition of an actuator (such as linear actuators or D.C. motors). The action layer
\nperformed the task of Central Pattern Generators (CPG), which produce rhythmic
\npatterns of activity. In particular, different biped and quadruped gait patterns were
\ncreated. An original neural model was specifically developed for assisting in the
\ncreation of these time-based patterns.
\nIt is shown in the thesis that Artificial Reflexes and CPGs can be configured
\nsuccessfully using this technique. The Artificial Reflexes were better at
\ngeneralising across different actuators, without changes, than traditional
\ncontrollers. Gaits such as pace, trot, gallop and pronk were successfully created
\nusing the CPGs. Experiments were conducted to determine whether modularity in
\nthe networks had an impact. It has been demonstrated that the degree of
\nmodularization in the network influences its evolvability, with more modular
\nnetworks evolving more efficiently.