A Closer Look at Invalid Action Masking in Policy Gradient Algorithms

TL;DR

This paper investigates the practice of invalid action masking in policy gradient algorithms for deep reinforcement learning, providing theoretical insights, empirical evidence of its importance, and analysis of different masking strategies.

Contribution

It offers a theoretical justification for invalid action masking, empirical demonstrations of its significance, and insights into various masking regimes in DRL.

Findings

Invalid action masking improves policy performance as invalid actions increase.

Removing masking after training can affect policy effectiveness.

Theoretical analysis supports the use of masking in policy gradient methods.

Abstract

In recent years, Deep Reinforcement Learning (DRL) algorithms have achieved state-of-the-art performance in many challenging strategy games. Because these games have complicated rules, an action sampled from the full discrete action distribution predicted by the learned policy is likely to be invalid according to the game rules (e.g., walking into a wall). The usual approach to deal with this problem in policy gradient algorithms is to "mask out" invalid actions and just sample from the set of valid actions. The implications of this process, however, remain under-investigated. In this paper, we 1) show theoretical justification for such a practice, 2) empirically demonstrate its importance as the space of invalid actions grows, and 3) provide further insights by evaluating different action masking regimes, such as removing masking after an agent has been trained using masking. The source code can be found at https://github.com/vwxyzjn/invalid-action-masking