ToyBox: Better Atari Environments for Testing Reinforcement Learning Agents

TL;DR

ToyBox offers reimplemented Atari environments with meaningful internal states, facilitating more effective testing and debugging of reinforcement learning agents beyond what traditional ALE environments provide.

Contribution

We introduce ToyBox, a set of Atari game reimplementations that provide semantically meaningful internal states for improved RL agent testing.

Findings

ToyBox enables more robust testing of RL agents.

ToyBox improves interpretability of agent behaviors.

ToyBox surpasses ALE in supporting model introspection.

Abstract

It is a widely accepted principle that software without tests has bugs. Testing reinforcement learning agents is especially difficult because of the stochastic nature of both agents and environments, the complexity of state-of-the-art models, and the sequential nature of their predictions. Recently, the Arcade Learning Environment (ALE) has become one of the most widely used benchmark suites for deep learning research, and state-of-the-art Reinforcement Learning (RL) agents have been shown to routinely equal or exceed human performance on many ALE tasks. Since ALE is based on emulation of original Atari games, the environment does not provide semantically meaningful representations of internal game state. This means that ALE has limited utility as an environment for supporting testing or model introspection. We propose ToyBox, a collection of reimplementations of these games that solves this critical problem and enables robust testing of RL agents.