AAAI Publications, Workshops at the Twenty-Fifth AAAI Conference on Artificial Intelligence

Font Size: 
Incremental Sensorimotor Learning with Constant Update Complexity
Arjan Gijsberts, Giorgio Metta

Last modified: 2011-08-24

Abstract


The robotics domain is challenging from a learning perspective, since subsequent observations are dependent and the environment is typically non-stationary. Successful modeling of sensorimotor relationships therefore necessitates an open-ended learning process that continuously updates existing models when novel observations become available, while at the same time respecting strict timing constraints. These requirements can be met by combining standard Bayesian regression with an exact update rule for incremental operation and a kernel approximation for non-linearity. The resulting method is characterized by a constant update complexity, which effectively allows lifelong operation. Furthermore, an experimental validation on predicting inverse dynamics of the iCub humanoid demonstrates superior generalization and timing performance with respect to competitive methods.

Full Text: PDF