Exploration in deep reinforcement learning: A survey

This paper reviews exploration techniques in deep reinforcement learning. Exploration are of primary importance when solving sparse reward problems. In problems, the is rare, which means that agent will not find often by acting randomly. such a scenario, it challenging for learning to learn rewards and actions association. Thus more sophisticated methods need be devised. review provides comprehensive overview existing approaches, categorised based on key contributions as: novel states, diverse behaviours, goal-based methods, probabilistic imitation-based safe random-based methods. Then, unsolved challenges discussed provide valuable future research directions. Finally, approaches different categories compared terms complexity, computational effort overall performance.