Growing Event Memories for Autonomous Robots

The aim of this paper is to provide a systematic overview of methods that are potentially useful for growing event memories in autonomous robots. Recognizing events in sequences of sensor and motor signals is viewed as an important step toward intelligent robot behaviour and part of concept formation, needed for learning tasks such as navigation, planning or language use. Here we concentrate on selfor unsupervised techniques in the context of autonomous, behaviour-based robots. Our ultimate goal will be to use information in timeseries of sensor and motor signals to recognize important events or characteristic features in the environment of a navigating mobile robot. We discuss potential approaches and their respective suitability for this problem and propose the basics for an event detection and memory architecture for the mobile robot platform KURT2.