Equilibrium-Driven Adaptive Behavior Design

Abstract

In autonomous robotics, scalability is a primary discriminator for evaluating a behavior design methodology. Such a proposed methodology must also allow efficient and effective conversion from desired to implemented behavior. From the concepts of equilibrium and homeostasis, it follows that behavior could be seen as driven rather than controlled. Homeostatic variables allow the development of need elements to completely implement drive and processing elements in a synthetic nervous system. Furthermore, an autonomous robot or system must act with a sense of meaning as opposed to being a human-command executor. Learning is fundamental in adding adaptability, and its efficient implementation will directly improve scalability. It is shown how using classical conditioning to learn obstacle avoidance can be implemented with need elements instead of an existing artificial neural network (ANN) solution.