MineDojo vs. MineRL: Empirical Evaluation

While working with friends from ML@B on a project of reinforcement learning (RL) and natural language processing (NLP) for goal-conditioned planning, one side task was choosing the right framework to test our models. Minecraft showed great potential of encapsulating great complexity while still being a widely controllable environment, so we chose it as our target setup. As of today there are two main frameworks that integrate Minecraft as an RL environment, namely MineRL and MineDojo. This article shows an empirical comparison of both frameworks on their execution performance.