Revisiting Model-based Value Expansion
This work addresses a practical limitation in reinforcement learning for researchers, providing insights into model-based methods but is incremental as it focuses on understanding existing failures rather than proposing new solutions.
The paper investigates why model-based value expansion methods underperform compared to simpler Dyna-style algorithms in reinforcement learning, attributing the failure to compounding model errors through empirical analysis using GPU-based physics simulators.
Model-based value expansion methods promise to improve the quality of value function targets and, thereby, the effectiveness of value function learning. However, to date, these methods are being outperformed by Dyna-style algorithms with conceptually simpler 1-step value function targets. This shows that in practice, the theoretical justification of value expansion does not seem to hold. We provide a thorough empirical study to shed light on the causes of failure of value expansion methods in practice which is believed to be the compounding model error. By leveraging GPU based physics simulators, we are able to efficiently use the true dynamics for analysis inside the model-based reinforcement learning loop. Performing extensive comparisons between true and learned dynamics sheds light into this black box. This paper provides a better understanding of the actual problems in value expansion. We provide future directions of research by empirically testing the maximum theoretical performance of current approaches.