PUB - Publikationen an der Universität Bielefeld

The design of a system that bootstraps an open-ended development is one of the most intriguing questions in Developmental Robotics. Inspired by evolution we propose an incremental design. We start with a reactive system that provides task-unspecific interaction with the environment. We extend this system by a multi-modal expectation generation system that learns and evaluates the regularities of the system-environment interaction. The two systems are coupled by means of an active resolution of the expectation mismatch. Such an extension allows the transition from reactive behavior to hypothesis testing and goal directed behavior. The proposed architecture is validated on the example of multi-modal learning and evaluation of auditory labels tested on humanoid robot ASIMO.