Maximum Entropy Gain Exploration for Long Horizon Multi-goal Reinforcement Learning
Silviu Pitis, Harris Chan, Stephen Zhao, Bradly Stadie, Jimmy Ba
Abstract
What goals should a multi-goal reinforcement learning agent pursue during training in long-horizon tasks? When the desired (test time) goal distribution is too distant to offer a useful learning signal, we argue that the agent should not pursue unobtainable goals. Instead, it should set its own intrinsic goals that maximize the entropy of the historical achieved goal distribution. We propose to optimize this objective by having the agent pursue past achieved goals in sparsely explored areas of the goal space, which focuses exploration on the frontier of the achievable goal set. We show that our strategy achieves an order of magnitude better sample efficiency than the prior state of the art on long-horizon multi-goal tasks including maze navigation and block stacking.
BibTeX
@InProceedings{pmlr-v119-pitis20a,
title = {Maximum Entropy Gain Exploration for Long Horizon Multi-goal Reinforcement Learning},
author = {Pitis, Silviu and Chan, Harris and Zhao, Stephen and Stadie, Bradly and Ba, Jimmy},
booktitle = {Proceedings of the 37th International Conference on Machine Learning},
pages = {7750--7761},
year = {2020},
editor = {III, Hal Daumé and Singh, Aarti},
volume = {119},
series = {Proceedings of Machine Learning Research},
month = {13--18 Jul},
publisher = {PMLR},
pdf = {http://proceedings.mlr.press/v119/pitis20a/pitis20a.pdf},
url = {https://proceedings.mlr.press/v119/pitis20a.html},
abstract = {What goals should a multi-goal reinforcement learning agent pursue during training in long-horizon tasks? When the desired (test time) goal distribution is too distant to offer a useful learning signal, we argue that the agent should not pursue unobtainable goals. Instead, it should set its own intrinsic goals that maximize the entropy of the historical achieved goal distribution. We propose to optimize this objective by having the agent pursue past achieved goals in sparsely explored areas of the goal space, which focuses exploration on the frontier of the achievable goal set. We show that our strategy achieves an order of magnitude better sample efficiency than the prior state of the art on long-horizon multi-goal tasks including maze navigation and block stacking.}
}