Incremental category learning in a real world artifact using growing dynamic cell structures

摘要

A biological or artifical agent operating in the real world must have means to categorise its environment. Since the real world is dynamic the problem arises how an agent can adaptively categorise non-stationary input data. In this paper we address this stability-flexibility trade-off using a mobile robot that has to solve a collecting task in a changing environment. We demonstrate that by using a Growing Dynamical Cell Structures algorithm the robot can incrementally learn categories.