Efficient Risk-Averse Reinforcement Learning

TL;DR

This paper introduces CeSoR, a novel risk-averse reinforcement learning method that combines a soft risk mechanism with a cross-entropy sampling module, improving sample efficiency and overcoming local optima in risk-sensitive tasks.

Contribution

It proposes a new CeSoR algorithm that separates risk sampling from policy optimization, enhancing risk aversion and efficiency in various RL benchmarks.

Findings

CeSoR outperforms standard risk-averse methods in maze navigation and autonomous driving.

The soft risk mechanism helps bypass local optima barriers.

Cross entropy sampling improves sample efficiency and risk handling.

Abstract

In risk-averse reinforcement learning (RL), the goal is to optimize some risk measure of the returns. A risk measure often focuses on the worst returns out of the agent's experience. As a result, standard methods for risk-averse RL often ignore high-return strategies. We prove that under certain conditions this inevitably leads to a local-optimum barrier, and propose a soft risk mechanism to bypass it. We also devise a novel Cross Entropy module for risk sampling, which (1) preserves risk aversion despite the soft risk; (2) independently improves sample efficiency. By separating the risk aversion of the sampler and the optimizer, we can sample episodes with poor conditions, yet optimize with respect to successful strategies. We combine these two concepts in CeSoR - Cross-entropy Soft-Risk optimization algorithm - which can be applied on top of any risk-averse policy gradient (PG) method. We demonstrate improved risk aversion in maze navigation, autonomous driving, and resource allocation benchmarks, including in scenarios where standard risk-averse PG completely fails.