TY - GEN
T1 - Obstacle tower
T2 - 28th International Joint Conference on Artificial Intelligence, IJCAI 2019
AU - Juliani, Arthur
AU - Khalifa, Ahmed
AU - Berges, Vincent Pierre
AU - Harper, Jonathan
AU - Teng, Ervin
AU - Henry, Hunter
AU - Crespi, Adam
AU - Togelius, Julian
AU - Lange, Danny
N1 - Funding Information:
The authors acknowledge the financial support from NSF grant (Award number 1717324 - ”RI: Small: General Intelligence through Algorithm Invention and Selection.”).
Funding Information:
The authors acknowledge the financial support from NSF grant (Award number 1717324 -?RI: Small: General Intelligence through Algorithm Invention and Selection.?). We would additionally like to thank Leon Chen, Jeff Shih, Marwan Mattar, Vilmantas Balasevicius, and Yuan Gao for helpful feedback and support during the development and evaluation of this environment, as well as all the participants who took part in the human performance evaluation process.
Publisher Copyright:
© 2019 International Joint Conferences on Artificial Intelligence. All rights reserved.
PY - 2019
Y1 - 2019
N2 - The rapid pace of recent research in AI has been driven in part by the presence of fast and challenging simulation environments. These environments often take the form of games; with tasks ranging from simple board games, to competitive video games. We propose a new benchmark - Obstacle Tower: a high fidelity, 3D, 3rd person, procedurally generated environment 1. An agent playing Obstacle Tower must learn to solve both low-level control and high-level planning problems in tandem while learning from pixels and a sparse reward signal. Unlike other benchmarks such as the Arcade Learning Environment, evaluation of agent performance in Obstacle Tower is based on an agent's ability to perform well on unseen instances of the environment. In this paper we outline the environment and provide a set of baseline results produced by current state-of-the-art Deep RL methods as well as human players. These algorithms fail to produce agents capable of performing near human level.
AB - The rapid pace of recent research in AI has been driven in part by the presence of fast and challenging simulation environments. These environments often take the form of games; with tasks ranging from simple board games, to competitive video games. We propose a new benchmark - Obstacle Tower: a high fidelity, 3D, 3rd person, procedurally generated environment 1. An agent playing Obstacle Tower must learn to solve both low-level control and high-level planning problems in tandem while learning from pixels and a sparse reward signal. Unlike other benchmarks such as the Arcade Learning Environment, evaluation of agent performance in Obstacle Tower is based on an agent's ability to perform well on unseen instances of the environment. In this paper we outline the environment and provide a set of baseline results produced by current state-of-the-art Deep RL methods as well as human players. These algorithms fail to produce agents capable of performing near human level.
UR - http://www.scopus.com/inward/record.url?scp=85074929352&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85074929352&partnerID=8YFLogxK
U2 - 10.24963/ijcai.2019/373
DO - 10.24963/ijcai.2019/373
M3 - Conference contribution
AN - SCOPUS:85074929352
T3 - IJCAI International Joint Conference on Artificial Intelligence
SP - 2684
EP - 2691
BT - Proceedings of the 28th International Joint Conference on Artificial Intelligence, IJCAI 2019
A2 - Kraus, Sarit
PB - International Joint Conferences on Artificial Intelligence
Y2 - 10 August 2019 through 16 August 2019
ER -