It is rational to adopt the average reward reinforcement learning algorithms for solving the absorbing goal states cyclical tasks: It has the merit of converging quickly and robustly.