The NetHack Learning Environment
- Küttler, Heinrich; Nardelli, Nantas; Raileanu, Roberta*; Selvatici, Marco; Grefenstette, Edward ; Rocktäschel, Tim
- Accepted abstract
[Join poster session]
Poster session from 15:00 to 16:00 EAT and from 20:45 to 21:45 EAT
Obtain the zoom password from ICLR
Hard and diverse tasks in complex environments drive progress in Reinforcement Learning (RL) research. To this end, we present the NetHack Learning Environment, a scalable, procedurally generated, rich and challenging environment for RL research based on the popular single-player terminal-based roguelike NetHack game, along with a suite of initial tasks. This environment is sufficiently complex to drive long-term research on exploration, planning, skill acquisition and complex policy learning, while dramatically reducing the computational resources required to gather a large amount of experience. We compare this environment and task suite to existing alternatives, and discuss why it is an ideal medium for testing the robustness and systematic generalization of RL agents. We demonstrate empirical success for early stages of the game using a distributed deep RL baseline and present a comprehensive qualitative analysis of agents trained in the environment.