MineRL BASALT

MineRL BASALT is a benchmark competition focused on advancing research in reinforcement learning agents that solve complex, ill-defined tasks within the Minecraft environment using human feedback. Current research emphasizes learning from human feedback (LfHF) to overcome the challenges of sparse rewards and ambiguous task specifications, employing diverse approaches including hierarchical reinforcement learning and imitation learning. The competition and its associated datasets (like BEDD) provide a standardized platform for evaluating algorithm performance and fostering the development of more robust and sample-efficient AI agents capable of tackling real-world problems with similar complexities. This work contributes significantly to the broader field of AI by pushing the boundaries of LfHF and providing valuable resources for the research community.