S-learning: a reinforcement learning method without parameter tuning

Discussed issues raised in two reinforcement learning algorithms, adaptive heuristic critic (AHC) and Q-learning. Aspects considered are convergence, parameter tuning, over-training, and computational and storage efficiency. Two new reinforcement learning mechanisms are proposed: S-learning (step-learning) and S&AHC learning. Particularly, the representation of the final cost map formed by S-learning series can be explicitly interpreted as the number of minimum movements to the goal state from each individual state. In addition, an adaptive (incremental) S-learning was proposed which incorporated S-learning and the technique of incremental learning to facilitate the practical implementation of neural reinforcement learning. All of S-learning series showed promising performances in exploring Sutton´s task (1991) of navigating in a maze.