Solving Richly Constrained Reinforcement Learning through State Augmentation and Reward Penalties

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Jiang, Hao, Mai, Tien, Varakantham, Pradeep, Hoang, Minh Huy
Format: Preprint
Veröffentlicht: 2023
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866913194063167488
author Jiang, Hao
Mai, Tien
Varakantham, Pradeep
Hoang, Minh Huy
author_facet Jiang, Hao
Mai, Tien
Varakantham, Pradeep
Hoang, Minh Huy
contents Constrained Reinforcement Learning has been employed to enforce safety constraints on policy through the use of expected cost constraints. The key challenge is in handling expected cost accumulated using the policy and not just in a single step. Existing methods have developed innovative ways of converting this cost constraint over entire policy to constraints over local decisions (at each time step). While such approaches have provided good solutions with regards to objective, they can either be overly aggressive or conservative with respect to costs. This is owing to use of estimates for "future" or "backward" costs in local cost constraints. To that end, we provide an equivalent unconstrained formulation to constrained RL that has an augmented state space and reward penalties. This intuitive formulation is general and has interesting theoretical properties. More importantly, this provides a new paradigm for solving constrained RL problems effectively. As we show in our experimental results, we are able to outperform leading approaches on multiple benchmark problems from literature.
format Preprint
id arxiv_https___arxiv_org_abs_2301_11592
institution arXiv
publishDate 2023
record_format arxiv
spellingShingle Solving Richly Constrained Reinforcement Learning through State Augmentation and Reward Penalties
Jiang, Hao
Mai, Tien
Varakantham, Pradeep
Hoang, Minh Huy
Machine Learning
Constrained Reinforcement Learning has been employed to enforce safety constraints on policy through the use of expected cost constraints. The key challenge is in handling expected cost accumulated using the policy and not just in a single step. Existing methods have developed innovative ways of converting this cost constraint over entire policy to constraints over local decisions (at each time step). While such approaches have provided good solutions with regards to objective, they can either be overly aggressive or conservative with respect to costs. This is owing to use of estimates for "future" or "backward" costs in local cost constraints. To that end, we provide an equivalent unconstrained formulation to constrained RL that has an augmented state space and reward penalties. This intuitive formulation is general and has interesting theoretical properties. More importantly, this provides a new paradigm for solving constrained RL problems effectively. As we show in our experimental results, we are able to outperform leading approaches on multiple benchmark problems from literature.
title Solving Richly Constrained Reinforcement Learning through State Augmentation and Reward Penalties
topic Machine Learning
url https://arxiv.org/abs/2301.11592