Addressing Non-Stationarity in Reinforcement Learning through Adaptive Exploration and Distribution Monitoring

Reinforcement learning (RL) algorithms often operate under the assumption of a stationary environment. However, in real-world applications, this assumption frequently fails. This paper investigates the impact of non-stationary environments on RL performance and proposes a novel approach that combines adaptive exploration strategies with distribution monitoring techniques to mitigate the effects of environment drift. Specifically, we integrate an intrinsic reward signal based on state novelty with a statistical test for detecting changes in the state distribution, triggering adjustments in the exploration-exploitation balance. Empirical results demonstrate the effectiveness of our approach in maintaining robust performance in dynamic environments.