Generalising Discrete Action Spaces with Conditional Action Trees

TL;DR

This paper introduces Conditional Action Trees, a structured approach to handle large and complex discrete action spaces in reinforcement learning, enabling generalization and reduction of action complexity.

Contribution

It proposes Conditional Action Trees as a novel method to structure and decompose action spaces in RL, facilitating generalization and multi-stage decision making.

Findings

Validated on environments with basic discrete actions

Effective in large combinatorial action spaces in RTS games

Reduces action space complexity significantly

Abstract

There are relatively few conventions followed in reinforcement learning (RL) environments to structure the action spaces. As a consequence the application of RL algorithms to tasks with large action spaces with multiple components require additional effort to adjust to different formats. In this paper we introduce {\em Conditional Action Trees} with two main objectives: (1) as a method of structuring action spaces in RL to generalise across several action space specifications, and (2) to formalise a process to significantly reduce the action space by decomposing it into multiple sub-spaces, favoring a multi-staged decision making approach. We show several proof-of-concept experiments validating our scheme, ranging from environments with basic discrete action spaces to those with large combinatorial action spaces commonly found in RTS-style games.