TY - GEN
T1 - Grounding Language to Entities and Dynamics for Generalization in Reinforcement Learning
AU - Hanjie, Austin W.
AU - Zhong, Victor
AU - Narasimhan, Karthik
N1 - Funding Information:
We are grateful to Ameet Deshpande, Jens Tuyls, Michael Hu, Shunyu Yao, Tsung-Yen Yang, Willie Chang and anonymous reviewers for their helpful comments and suggestions. We would also like to thank the anonymous AMT workers for their indispensable contributions to this work. This work was financially supported by the Princeton SEAS Senior Thesis Fund.
Publisher Copyright:
Copyright © 2021 by the author(s)
PY - 2021
Y1 - 2021
N2 - We investigate the use of natural language to drive the generalization of control policies and introduce the new multi-task environment MESSENGER with free-form text manuals describing the environment dynamics. Unlike previous work, MESSENGER does not assume prior knowledge connecting text and state observations - the control policy must simultaneously ground the game manual to entity symbols and dynamics in the environment. We develop a new model, EMMA (Entity Mapper with Multi-modal Attention) which uses an entity-conditioned attention module that allows for selective focus over relevant descriptions in the manual for each entity in the environment. EMMA is end-to-end differentiable and learns a latent grounding of entities and dynamics from text to observations using only environment rewards. EMMA achieves successful zero-shot generalization to unseen games with new dynamics, obtaining a 40% higher win rate compared to multiple baselines. However, win rate on the hardest stage of MESSENGER remains low (10%), demonstrating the need for additional work in this direction.
AB - We investigate the use of natural language to drive the generalization of control policies and introduce the new multi-task environment MESSENGER with free-form text manuals describing the environment dynamics. Unlike previous work, MESSENGER does not assume prior knowledge connecting text and state observations - the control policy must simultaneously ground the game manual to entity symbols and dynamics in the environment. We develop a new model, EMMA (Entity Mapper with Multi-modal Attention) which uses an entity-conditioned attention module that allows for selective focus over relevant descriptions in the manual for each entity in the environment. EMMA is end-to-end differentiable and learns a latent grounding of entities and dynamics from text to observations using only environment rewards. EMMA achieves successful zero-shot generalization to unseen games with new dynamics, obtaining a 40% higher win rate compared to multiple baselines. However, win rate on the hardest stage of MESSENGER remains low (10%), demonstrating the need for additional work in this direction.
UR - http://www.scopus.com/inward/record.url?scp=85161306522&partnerID=8YFLogxK
UR - http://www.scopus.com/inward/citedby.url?scp=85161306522&partnerID=8YFLogxK
M3 - Conference contribution
AN - SCOPUS:85161306522
T3 - Proceedings of Machine Learning Research
SP - 4051
EP - 4062
BT - Proceedings of the 38th International Conference on Machine Learning, ICML 2021
PB - ML Research Press
T2 - 38th International Conference on Machine Learning, ICML 2021
Y2 - 18 July 2021 through 24 July 2021
ER -