Objective Mismatch in Model-based Reinforcement Learning

Download the paper!Read on ArXiv!Run the code!Video available!
ABSTRACT:
hide & show
Model-based reinforcement learning (MBRL) is a powerful framework for data-efficiently learning control of continuous tasks. Recent work in MBRL has mostly focused on using more advanced function approximators and planning schemes, with little development of the general framework. In this paper, we identify a fundamental issue of the standard MBRL framework – what we call objective mismatch. Objective mismatch arises when one objective is optimized in the hope that a second, often uncorrelated, metric will also be optimized. In the context of MBRL, we characterize the objective mismatch between training the forward dynamics model w.r.t. the likelihood of the one-step ahead prediction, and the overall goal of improving performance on a downstream control task. For example, this issue can emerge with the realization that dynamics models effective for a specific task do not necessarily need to be globally accurate, and vice versa globally accurate models might not be sufficiently accurate locally to obtain good control performance on a specific task. In our experiments, we study this objective mismatch issue and demonstrate that the likelihood of one-step ahead predictions is not always correlated with control performance. This observation highlights a critical limitation in the MBRL framework which will require further research to be fully understood and addressed. We propose an initial method to mitigate the mismatch issue by re-weighting dynamics model training. Building on it, we conclude with a discussion about other potential directions of research for addressing this issue.

What you need to know:

  1. Training an accurate dynamics model in MBRL does not guarantee task performance will be high with that controller.
  2. A whole new spectrum of model-based control algorithms is needed to better link modeling to performance (hard open problem).

Citation

@article{lambert2020objective,
 title={Objective Mismatch in Model-based Reinforcement Learning},
 author={Lambert, Nathan and Amos, Brandon and Yadan, Omry and Calandra, Roberto},
 journal={Learning for Decision and Control (L4DC)},
 year={2020}
}