On the Geometry of Reinforcement Learning in Continuous State and Action Spaces

Fuente: arXiv
Saved in:
Bibliographic Details
Main Authors: Tiwari, Saket, Gottesman, Omer, Konidaris, George
Format: Preprint
Published: 2022
Subjects:
Online Access:
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866910561886797824
author Tiwari, Saket
Gottesman, Omer
Konidaris, George
author_facet Tiwari, Saket
Gottesman, Omer
Konidaris, George
contents Advances in reinforcement learning have led to its successful application in complex tasks with continuous state and action spaces. Despite these advances in practice, most theoretical work pertains to finite state and action spaces. We propose building a theoretical understanding of continuous state and action spaces by employing a geometric lens. Central to our work is the idea that the transition dynamics induce a low dimensional manifold of reachable states embedded in the high-dimensional nominal state space. We prove that, under certain conditions, the dimensionality of this manifold is at most the dimensionality of the action space plus one. This is the first result of its kind, linking the geometry of the state space to the dimensionality of the action space. We empirically corroborate this upper bound for four MuJoCo environments. We further demonstrate the applicability of our result by learning a policy in this low dimensional representation. To do so we introduce an algorithm that learns a mapping to a low dimensional representation, as a narrow hidden layer of a deep neural network, in tandem with the policy using DDPG. Our experiments show that a policy learnt this way perform on par or better for four MuJoCo control suite tasks.
format Preprint
id arxiv_https___arxiv_org_abs_2301_00009
institution arXiv
publishDate 2022
record_format arxiv
spellingShingle On the Geometry of Reinforcement Learning in Continuous State and Action Spaces
Tiwari, Saket
Gottesman, Omer
Konidaris, George
Machine Learning
Artificial Intelligence
Advances in reinforcement learning have led to its successful application in complex tasks with continuous state and action spaces. Despite these advances in practice, most theoretical work pertains to finite state and action spaces. We propose building a theoretical understanding of continuous state and action spaces by employing a geometric lens. Central to our work is the idea that the transition dynamics induce a low dimensional manifold of reachable states embedded in the high-dimensional nominal state space. We prove that, under certain conditions, the dimensionality of this manifold is at most the dimensionality of the action space plus one. This is the first result of its kind, linking the geometry of the state space to the dimensionality of the action space. We empirically corroborate this upper bound for four MuJoCo environments. We further demonstrate the applicability of our result by learning a policy in this low dimensional representation. To do so we introduce an algorithm that learns a mapping to a low dimensional representation, as a narrow hidden layer of a deep neural network, in tandem with the policy using DDPG. Our experiments show that a policy learnt this way perform on par or better for four MuJoCo control suite tasks.
title On the Geometry of Reinforcement Learning in Continuous State and Action Spaces
topic Machine Learning
Artificial Intelligence
url https://arxiv.org/abs/2301.00009