Sim-Env: Decoupling OpenAI Gym Environments from Simulation Models

Reinforcement learning (RL) is being used to create self-adaptive agents, where RL researchers commonly create and employ simulations of the problem domains to train and evaluate various RL algorithms and their variants. This activity is in need of methodological and tool-based support, particularly concerning the reuse of model-and simulation-related code across various RL experiments. We propose a workflow and tool for the decoupled development and maintenance of multi-purpose agent-based models and derived single-purpose RL environments, enabling the researcher to swap out environments with ones representing different perspectives or different reward models, all while keeping the underlying domain model intact and separate.