Hierarchical Attentive Multiple Models for Execution and Recognition of Actions through Training Inverse and Forward Models Explorationally

We present an extension to Demiris’ and Khadhouri’s Hierarchical Attentive Multiple Models for Execution and Recognition of actions, which is more honest to the developmental paradigm of robotics. Specifically, we replace their human-coded prediction models with ones that the robot learns themselves through exploratory motor babbling guided by Categorical Intrinsic-Based Motivation.