Skip to Main Content
This paper addresses a broad spectrum of machine learning problems. Actions by embodied agents automatically generate training data for the learning mechanisms, so that a humanoid robot develops categorization autonomously. Cognitive capabilities of the humanoid robot are developmentally created, starting from abilities for detecting, segmenting, and recognizing objects. Such mature abilities are integrated with the deeper developmental learning mechanisms required to create those abilities out of the robot's physical experiences. This work presents strategies for learning task sequences and to recognize objects employed on such tasks from human-robot interaction cues. Learning strategies are also presented for the control of both oscillatory and non-oscillatory movements for the execution of these learned tasks. Self-exploration of the world automatically introduces the robot to new training data.