학술논문
Insights From the NeurIPS 2021 NetHack Challenge
Document Type
Working Paper
Author
Hambro, Eric; Mohanty, Sharada; Babaev, Dmitrii; Byeon, Minwoo; Chakraborty, Dipam; Grefenstette, Edward; Jiang, Minqi; Jo, Daejin; Kanervisto, Anssi; Kim, Jongmin; Kim, Sungwoong; Kirk, Robert; Kurin, Vitaly; Küttler, Heinrich; Kwon, Taehwon; Lee, Donghoon; Mella, Vegard; Nardelli, Nantas; Nazarov, Ivan; Ovsov, Nikita; Parker-Holder, Jack; Raileanu, Roberta; Ramanauskas, Karolis; Rocktäschel, Tim; Rothermel, Danielle; Samvelyan, Mikayel; Sorokin, Dmitry; Sypetkowski, Maciej; Sypetkowski, Michał
Source
Subject
Language
Abstract
In this report, we summarize the takeaways from the first NeurIPS 2021 NetHack Challenge. Participants were tasked with developing a program or agent that can win (i.e., 'ascend' in) the popular dungeon-crawler game of NetHack by interacting with the NetHack Learning Environment (NLE), a scalable, procedurally generated, and challenging Gym environment for reinforcement learning (RL). The challenge showcased community-driven progress in AI with many diverse approaches significantly beating the previously best results on NetHack. Furthermore, it served as a direct comparison between neural (e.g., deep RL) and symbolic AI, as well as hybrid systems, demonstrating that on NetHack symbolic bots currently outperform deep RL by a large margin. Lastly, no agent got close to winning the game, illustrating NetHack's suitability as a long-term benchmark for AI research.
Comment: Under review at PMLR for the NeuRIPS 2021 Competition Workshop Track, 10 pages + 10 in appendices
Comment: Under review at PMLR for the NeuRIPS 2021 Competition Workshop Track, 10 pages + 10 in appendices