Reward bonuses with gain scheduling inspired by iterative deepening search

This paper introduces a novel method of adding intrinsic bonuses to task-oriented reward function in order efficiently facilitate reinforcement learning search. While various have been designed date, they are analogous the depth-first and breadth-first search algorithms graph theory. paper, therefore, first designs two for each them. Then, heuristic gain scheduling is applied bonuses, inspired by iterative deepening search, which known inherit advantages algorithms. The proposed expected allow agent reach best solution deeper states gradually exploring unknown states. In three locomotion tasks with dense rewards simple sparse rewards, it shown that types contribute performance improvement different complementarily. addition, combining them scheduling, all can be accomplished high performance.