• Classified by Topic • Classified by Publication Type • Sorted by Date • Sorted by First Author Last Name • Classified by Funding Source •
f-Policy Gradients: A General Framework for Goal Conditioned RL using f-Divergences.
Siddhant Agarwal, Ishan
Durugkar, Peter Stone, and Amy Zhang.
In Conference on Neural Information
Processing Systems, December 2023.
[PDF]3.4MB [slides.pptx]13.6MB [poster.pdf]1.8MB
Goal-Conditioned Reinforcement Learning (RL) problems often have access tosparse rewards where the agent receives a reward signal only when it hasachieved the goal, making policy optimization a difficult problem. Several worksaugment this sparse reward with a learned dense reward function, but this canlead to sub-optimal policies if the reward is misaligned. Moreover, recent workshave demonstrated that effective shaping rewards for a particular problem candepend on the underlying learning algorithm. This paper introduces a novel wayto encourage exploration called f-Policy Gradients, or f-PG. f-PG minimizes thef-divergence between the agent’s state visitation distribution and the goal,which we show can lead to an optimal policy. We derive gradients for variousf-divergences to optimize this objective. Our learning paradigm provides denselearning signals for exploration in sparse reward settings. We further introducean entropy-regularized policy optimization objective, that we call state-MaxEntRL (or s-MaxEnt RL) as a special case of our objective. We show that severalmetric-based shaping rewards like L2 can be used with s-MaxEnt RL, providing acommon ground to study such metric-based shaping rewards with efficientexploration. We find that f-PG has better performance compared to standardpolicy gradient methods on a challenging gridworld as well as the Point Maze andFetchReach environments. More information on our website
@InProceedings{agarwal2023fpg,
author = {Siddhant Agarwal and Ishan Durugkar and Peter Stone and Amy Zhang},
title = {f-Policy Gradients: A General Framework for Goal Conditioned RL using f-Divergences},
booktitle = {Conference on Neural Information Processing Systems},
year = {2023},
month = {December},
location = {New Orleans},
abstract = {
Goal-Conditioned Reinforcement Learning (RL) problems often have access to
sparse rewards where the agent receives a reward signal only when it has
achieved the goal, making policy optimization a difficult problem. Several works
augment this sparse reward with a learned dense reward function, but this can
lead to sub-optimal policies if the reward is misaligned. Moreover, recent works
have demonstrated that effective shaping rewards for a particular problem can
depend on the underlying learning algorithm. This paper introduces a novel way
to encourage exploration called f-Policy Gradients, or f-PG. f-PG minimizes the
f-divergence between the agent’s state visitation distribution and the goal,
which we show can lead to an optimal policy. We derive gradients for various
f-divergences to optimize this objective. Our learning paradigm provides dense
learning signals for exploration in sparse reward settings. We further introduce
an entropy-regularized policy optimization objective, that we call state-MaxEnt
RL (or s-MaxEnt RL) as a special case of our objective. We show that several
metric-based shaping rewards like L2 can be used with s-MaxEnt RL, providing a
common ground to study such metric-based shaping rewards with efficient
exploration. We find that f-PG has better performance compared to standard
policy gradient methods on a challenging gridworld as well as the Point Maze and
FetchReach environments. More information on our website
},
}
Generated by bib2html.pl (written by Patrick Riley ) on Wed Apr 17, 2024 18:42:49