An incremental learning algorithm for autonomous mental development of robots

Autonomous mental development of robots should generate effective internal representations from limited experience, and learn incrementally. As a result, a robot´s learning strategy plays an important role in its life. This paper comes up with a new algorithm: incre-tree, which is a hierarchical method and only needs four parameters defined by the user. Incre-tree first processes some samples in a batch fashion and constructs an initial concept tree, then computes each new sample to update one leaf, the sample is discarded before the next one arrives. A leaf begins to divide into two parts when it contains a certain number of samples, thus the concept tree could grow continuously.