Safe Reinforcement Learning in Black-Box Environments via Adaptive Shielding

Fuente: arXiv
Gespeichert in:
Bibliographische Detailangaben
Hauptverfasser: Bethell, Daniel, Gerasimou, Simos, Calinescu, Radu, Imrie, Calum
Format: Preprint
Veröffentlicht: 2024
Schlagworte:
Online-Zugang:
Tags: Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
_version_ 1866912555262279680
author Bethell, Daniel
Gerasimou, Simos
Calinescu, Radu
Imrie, Calum
author_facet Bethell, Daniel
Gerasimou, Simos
Calinescu, Radu
Imrie, Calum
contents Empowering safe exploration of reinforcement learning (RL) agents during training is a critical challenge towards their deployment in many real-world scenarios. When prior knowledge of the domain or task is unavailable, training RL agents in unknown, black-box environments presents an even greater safety risk. We introduce ADVICE (Adaptive Shielding with a Contrastive Autoencoder), a novel post-shielding technique that distinguishes safe and unsafe features of state-action pairs during training, and uses this knowledge to protect the RL agent from executing actions that yield likely hazardous outcomes. Our comprehensive experimental evaluation against state-of-the-art safe RL exploration techniques shows that ADVICE significantly reduces safety violations (approx 50%) during training, with a competitive outcome reward compared to other techniques.
format Preprint
id arxiv_https___arxiv_org_abs_2405_18180
institution arXiv
publishDate 2024
record_format arxiv
spellingShingle Safe Reinforcement Learning in Black-Box Environments via Adaptive Shielding
Bethell, Daniel
Gerasimou, Simos
Calinescu, Radu
Imrie, Calum
Artificial Intelligence
Machine Learning
Empowering safe exploration of reinforcement learning (RL) agents during training is a critical challenge towards their deployment in many real-world scenarios. When prior knowledge of the domain or task is unavailable, training RL agents in unknown, black-box environments presents an even greater safety risk. We introduce ADVICE (Adaptive Shielding with a Contrastive Autoencoder), a novel post-shielding technique that distinguishes safe and unsafe features of state-action pairs during training, and uses this knowledge to protect the RL agent from executing actions that yield likely hazardous outcomes. Our comprehensive experimental evaluation against state-of-the-art safe RL exploration techniques shows that ADVICE significantly reduces safety violations (approx 50%) during training, with a competitive outcome reward compared to other techniques.
title Safe Reinforcement Learning in Black-Box Environments via Adaptive Shielding
topic Artificial Intelligence
Machine Learning
url https://arxiv.org/abs/2405.18180