Humanoid robot learning by demonstration based on visual bootstrapping technique

In this document a Programming by Demonstration proposal is described. It was applied to a humanoid robot in order to provide it with the ability to learn new high level behaviors. Particularly, a catching gesture-frequently used in soccer-was demonstrated by a person and learnt by a robot. The learning process was mainly based on a Genetic Algorithm (GA) and Principal Components Analysis (PCA) technique combination. While GA was used to support a more robust searching process, PCA helps it to be more efficient. Finally, the learnt behavior was executed on both simulated environment and real platform. Several results are shown and further work is discussed.