Intelligent problem-solving as integrated hierarchical reinforcement learning

According to cognitive psychology and related disciplines, the development of complex problem-solving behaviour in biological agents depends on hierarchical mechanisms. Hierarchical reinforcement learning is a promising computational approach that may eventually yield comparable artificial robots. However, so far, abilities many human non-human animals are clearly superior those systems. Here we propose steps integrate biologically inspired mechanisms enable advanced skills agents. We first review literature highlight importance compositional abstraction predictive processing. Then relate gained insights with contemporary methods. Interestingly, our results suggest all identified have been implemented individually isolated architectures, raising question why there exists no single unifying architecture integrates them. As final contribution, address this by providing an integrative perspective challenges develop such architecture. expect guide more sophisticated cognitively machine architectures. Although do well when rules rigid, as games, they fare poorly real-world scenarios where small changes environment or required actions can impair performance. The authors provide overview foundations problem-solving,