ToyBox: Better Atari Environments for Testing Reinforcement Learning Agents
This work provides a solution for researchers and developers in reinforcement learning who need better environments for testing and introspection, though it is incremental as it builds upon existing ALE benchmarks.
The paper tackled the problem of testing reinforcement learning agents by addressing the limitations of the Arcade Learning Environment (ALE), which lacks semantically meaningful game state representations, and proposed ToyBox, a collection of reimplemented Atari games that enables robust testing and model introspection.
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.