Goal-directed encoding of task knowledge for robotic skill acquisition

An intelligent control technique has been developed that integrates knowledge-based systems and artificial neural networks in order to emulate behavioral aspects of human skill acquisition. With strategies for learning and the capability to learn, the Robotic Skill Acquisition Architecture (RSA²) utilizes transitions between declarative and reflexive forms of processing to enable system adaptation and optimization. The robot learns through experience how to perfect tasks initially specified in a high-level task language. Knowledge-based systems encode neural network learning strategies, and skill acquisition is associated with the shift from a predominantly feedback-oriented, rule-based representation of control to a predominantly feedforward, network-based form. How rule-based goal-directed task descriptions are used to obtain initial `rough-cut´ system performance is described. The expressive and flexible nature of RSA² goals is demonstrated