The challenge served as a direct comparison between neural and symbolic AI, as well as hybrid systems, demonstrating that on NetHack symbolic bots currently outperform deep RL by a large margin.
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.
S. Mohanty
6 papers
Mikayel Samvelyan
7 papers
Nantas Nardelli
9 papers
A. Kanervisto
7 papers
Edward Grefenstette
14 papers
Heinrich Kuttler
4 papers
R. Raileanu
5 papers
Tim Rocktaschel
5 papers
Maciej Sypetkowski
2 papers
Jack Parker-Holder
4 papers
Robert Kirk
3 papers
Eric Hambro
5 papers
Vitaly Kurin
4 papers
Dipam Chakraborty
2 papers
Minqi Jiang
4 papers
Sungwoong Kim
2 papers
I. Nazarov
2 papers
DaeJin Jo
2 papers
Dmitrii Babaev
1 papers
Mi-Ra Byeon
1 papers
Jongmin Kim
1 papers
Taehwon Kwon
1 papers
Donghoon Lee
1 papers
Vegard Mella
2 papers
Nikita Ovsov
1 papers
Karolis Ramanauskas
1 papers
Dan Rothermel
2 papers
Dmitry Sorokin
1 papers
Michal Sypetkowski
1 papers