Answers are grounded in this paper and cite their sources.
Deep reinforcement learning (DRL) has been widely used in agent research across various video games, demonstrating its effectiveness. Recently, there has been increasing interest in DRL research in complex environments such as Roguelike games. Among them, the game NetHack has gained research attention. In this study, we aim to train a DRL agent for efficient learning with reduced training costs using the NetHack Learning Environment (NLE). A variational autoencoder (VAE) can efficiently transform input states into latent representations. Therefore, we propose a method utilizing VAE, considering it as a potentially efficient DRL model. Additionally, because the rewards provided by NLE are sparse, we train a DRL agent with additional rewards. As a result, we find that providing additional rewards to the agent is effective for tackling NLE, where the reward space is sparse. Furthermore, our results suggest that directly using the latent representation of the VAE as input to the DRL model can be effective.
No takes yet. Share an insight, caveat, or question.
Onuki et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: