Learning to adapt: A network and learning algorithm for flexible robotic control

One of the final hurdles to the widespread integration of robotic systems is the lack of true robotic autonomy. To this end a network structure, a Behaviour Network (BNet) and learning algorithm for the control of autonomous robotics is presented. The goal of this network learning algorithm is to enable flexible, adaptable control of autonomous systems, and in particular, robotic systems. The flexibility of the networks arises from its ability to learn online and adjust both architecture and function. Also, the network is designed with the goal of enveloping it within a three-fold approach to controller development in mind.