IRL with Partial Observations using the Principle of Uncertain Maximum Entropy

TL;DR

This paper extends the maximum entropy principle to scenarios with partial observations by introducing the principle of uncertain maximum entropy, which accounts for model dependency and improves robustness in inverse reinforcement learning.

Contribution

It proposes the principle of uncertain maximum entropy and an EM-based solution to handle partial observations in maximum entropy frameworks.

Findings

Enhanced robustness to noisy data in inverse reinforcement learning

Demonstrated effectiveness in a maximum causal entropy domain

Generalized from latent maximum entropy approach

Abstract

The principle of maximum entropy is a broadly applicable technique for computing a distribution with the least amount of information possible while constrained to match empirically estimated feature expectations. However, in many real-world applications that use noisy sensors computing the feature expectations may be challenging due to partial observation of the relevant model variables. For example, a robot performing apprenticeship learning may lose sight of the agent it is learning from due to environmental occlusion. We show that in generalizing the principle of maximum entropy to these types of scenarios we unavoidably introduce a dependency on the learned model to the empirical feature expectations. We introduce the principle of uncertain maximum entropy and present an expectation-maximization based solution generalized from the principle of latent maximum entropy. Finally, we experimentally demonstrate the improved robustness to noisy data offered by our technique in a maximum causal entropy inverse reinforcement learning domain.