Section: New Results
New algorithms for intrinsically motivated exploration
Based on the formal landscape described in the previous paragraph, a subspace of variants of knowledge-based predictive intrinsic motivation systems was investigated and systematically compared both based on measures of the self-organization of developmental trajectory and on measures of the efficiency of learning in simulated robotic sensorimotor spaces. Based on these systematic evaluations and comparisons, a new version of the Intelligent Adaptive Curiosity algorithm (IAC) have been developed which clearly outperforms the one described in  . It was tested in complicated simulated multi-arm forward model learning in robots and was shown to outperform random exploration, standard active learning heuristics, and previous version of IAC. The systematic studies of basic variants was presented in  .
Publications on the new IAC algorithm are under writing.