Objective Mismatch in Model-based Reinforcement Learning

Abstract

Model-based reinforcement learning (MBRL) has been shown to be 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 the objective mismatch issue. 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.

arXiv comment9 pages, 2 pages references, 5 pages appendices

References in corpus (11)

• Differentiable MPC for End-to-end Planning and Control
• SOLAR: Deep Structured Representations for Model-Based Reinforcement Learning
• Task-based End-to-end Model Learning in Stochastic Optimization
• Universal Planning Networks
• Exploring Model-based Planning with Policy Networks
• DeepMDP: Learning Continuous Latent Space Models for Representation Learning
• Robust Reinforcement Learning for Continuous Control with Model Misspecification
• Path Integral Networks: End-to-End Differentiable Optimal Control
• Goal-Driven Dynamics Learning via Bayesian Optimization
• Self-Correcting Models for Model-Based Reinforcement Learning
• Learning Dynamics Model in Reinforcement Learning by Incorporating the Long Term Future

Cited by in corpus (4)

• Trajectory Planning with Deep Reinforcement Learning in High-Level Action Spaces
• The Differentiable Cross-Entropy Method
• Learning MDPs from Features: Predict-Then-Optimize for Sequential Decision Problems by Reinforcement Learning
• Mitigating Value Hallucination in Dyna Planning via Multistep Predecessor Models