Value-Distributional Model-Based Reinforcement Learning
This work addresses uncertainty quantification for sequential decision-making in reinforcement learning, offering a novel method that improves performance in continuous-control domains, though it is incremental in building on distributional reinforcement learning.
The paper tackles the problem of quantifying uncertainty in policy performance by learning the posterior distribution over value functions from model-based Bayesian reinforcement learning, introducing a Bellman operator for value distribution functions and proposing the Epistemic Quantile-Regression (EQR) algorithm combined with soft actor-critic, which shows performance benefits across continuous-control tasks.
Quantifying uncertainty about a policy's long-term performance is important to solve sequential decision-making tasks. We study the problem from a model-based Bayesian reinforcement learning perspective, where the goal is to learn the posterior distribution over value functions induced by parameter (epistemic) uncertainty of the Markov decision process. Previous work restricts the analysis to a few moments of the distribution over values or imposes a particular distribution shape, e.g., Gaussians. Inspired by distributional reinforcement learning, we introduce a Bellman operator whose fixed-point is the value distribution function. Based on our theory, we propose Epistemic Quantile-Regression (EQR), a model-based algorithm that learns a value distribution function. We combine EQR with soft actor-critic (SAC) for policy optimization with an arbitrary differentiable objective function of the learned value distribution. Evaluation across several continuous-control tasks shows performance benefits with respect to both model-based and model-free algorithms. The code is available at https://github.com/boschresearch/dist-mbrl.