Constrained Policy Improvement for Safe and Efficient Reinforcement Learning

TL;DR

This paper introduces Rerouted Behavior Improvement (RBI), a policy improvement method for RL that accounts for Q-function errors, reducing negative policy shifts and enhancing safety and data efficiency in batch and iterative learning scenarios.

Contribution

The paper presents RBI, a novel policy improvement algorithm that mitigates the impact of Q-value estimation errors, improving safety and efficiency in reinforcement learning.

Findings

RBI avoids catastrophic performance degradation.

RBI increases data efficiency in high-variance actions.

RBI outperforms greedy and other constrained algorithms in experiments.

Abstract

We propose a policy improvement algorithm for Reinforcement Learning (RL) which is called Rerouted Behavior Improvement (RBI). RBI is designed to take into account the evaluation errors of the Q-function. Such errors are common in RL when learning the $Q$-value from finite past experience data. Greedy policies or even constrained policy optimization algorithms which ignore these errors may suffer from an improvement penalty (i.e. a negative policy improvement). To minimize the improvement penalty, the RBI idea is to attenuate rapid policy changes of low probability actions which were less frequently sampled. This approach is shown to avoid catastrophic performance degradation and reduce regret when learning from a batch of past experience. Through a two-armed bandit with Gaussian distributed rewards example, we show that it also increases data efficiency when the optimal action has a high variance. We evaluate RBI in two tasks in the Atari Learning Environment: (1) learning from observations of multiple behavior policies and (2) iterative RL. Our results demonstrate the advantage of RBI over greedy policies and other constrained policy optimization algorithms as a safe learning approach and as a general data efficient learning algorithm. An anonymous Github repository of our RBI implementation is found at https://github.com/eladsar/rbi.