Structured Multimodal Reinforcement Learning for Playing NetHack

Keisuke IZUMIYA

We improve the learning method of a reinforcement learning agent playing NetHack from three perspectives: inventory, in-game string, and the use of expert data.