Output details
11 - Computer Science and Informatics
University of Aberdeen
Learning regions for building a world model from clusters in probability distributions
<24>This is a new (and more natural) approach to bootstrapping a domain model, without prior knowledge (an alternative to Mugan and Kuipers’s QLAP). It is very significant for developmental robotics because the learning method is applicable recursively, at multiple levels within a cognitive robot; e.g. from the sensor level right up to finding regions in higher order relationships which define concepts. This work has been extended to two-dimensional regions in rules (paper in ICDL-EpiRob 2013, currently no DOI). ICDL-EpiRob is the premier conference for developmental robotics, joining the previous EpiRob (since 2001) and ICDL (since 2002) conferences.