On 'parts' and 'wholes' of adaptive behavior: functional modularity and diachronic structure in recurrent neural robot controllers

摘要

This paper discusses the issue of self-organized modularity for adaptive robot behavior. In particular we introduce the concept of functional modularity through diachronic structure in recurrent neural robot control mechanisms. We demonstrate how behavioral adaptivity through quasi-modularity at a functional level can be realized in monolithic recurrent neurocontrollers through context-dependent use of first- and higher-order feedback, i.e. without actual modules in the conventional sense.