Deep reinforcement learning in non-stationary environments
- Publication Type:
- Thesis
- Issue Date:
- 2024
Open Access
Copyright Clearance Process
- Recently Added
- In Progress
- Open Access
This item is open access.
Traditional deep reinforcement learning approaches can be characterized as the search for a policy that maximizes cumulative rewards in an unknown but stationary environment with fixed state transitions and reward functions. However, the stationarity assumption does not hold in many practical scenarios, such as outdoor robots encountering different terrains and lighting conditions. These non-stationary environments are subject to sudden and unpredictable change points that violate the stationarity assumption. In such cases, simply following the traditional deep reinforcement learning method will lead to performance degradation or even failure. To address this challenge, this thesis focuses on deep reinforcement learning in environments with time-varying unknown non-stationarity. We span the entire learning pipeline, formally define the problem and formally define the problem and clearly identify the similarities and differences with existing schemes. Specifically, our proposed methods have two inseparable processes: detecting environmental changes and adapting using the change information. The empirical results demonstrate better performance of our four methods relative to several current algorithms.
Please use this identifier to cite or link to this item: