Reward Centering
This incremental improvement addresses performance issues in reinforcement learning algorithms for continuing tasks, benefiting researchers and practitioners in the field.
The paper tackles the problem of improving discounted methods in continuing reinforcement learning by centering rewards through subtracting their empirical average, resulting in substantial performance improvements, especially as the discount factor approaches one, and making methods robust to reward shifts.
We show that discounted methods for solving continuing reinforcement learning problems can perform significantly better if they center their rewards by subtracting out the rewards' empirical average. The improvement is substantial at commonly used discount factors and increases further as the discount factor approaches one. In addition, we show that if a problem's rewards are shifted by a constant, then standard methods perform much worse, whereas methods with reward centering are unaffected. Estimating the average reward is straightforward in the on-policy setting; we propose a slightly more sophisticated method for the off-policy setting. Reward centering is a general idea, so we expect almost every reinforcement-learning algorithm to benefit by the addition of reward centering.