A new algorithm for reactive learning of obstacle avoidance

Abstract

Previous work in sensorymotor learning of reactive behaviour with Hierarchical Extended Self-Organizing Maps (HESOM ́s)) is examined. A number of observations of the strengths and weaknesses of such a system are outlined and several improvements are suggested. However, real robot learning has requirements that are only fullfiled with a new algorithm presented here. This algorithm is capable of very quick learning due to its dynamic structures. But, more interestingly, it deals more effectivelywith continuous learning anddifferent distributions of data samplings.