Self-Improving Safety Performance of Reinforcement Learning Based Driving with Black-Box Verification Algorithms

Resul Dagdanov*, Halil Durmus, Nazim Kemal Ure

*Corresponding author for this work

Research output: Chapter in Book/Report/Conference proceedingConference contributionpeer-review

1 Citation (Scopus)

Abstract

In this work, we propose a self-improving artificial intelligence system to enhance the safety performance of reinforcement learning (RL)-based autonomous driving (AD) agents using black-box verification methods. RL algorithms have become popular in AD applications in recent years. However, the performance of existing RL algorithms heavily depends on the diversity of training scenarios. A lack of safety-critical scenarios during the training phase could result in poor generalization performance in real-world driving applications. We propose a novel framework in which the weaknesses of the training set are explored through black-box verification methods. After discovering AD failure scenarios, the RL agent's training is re-initiated via transfer learning to improve the performance of previously unsafe scenarios. Simulation results demonstrate that our approach efficiently discovers safety failures of action decisions in RL-based adaptive cruise control (ACC) applications and significantly reduces the number of vehicle collisions through iterative applications of our method. The source code is publicly available at https://github.com/data-and-decision-lab/self-improving-RL.

Original languageEnglish
Title of host publicationProceedings - ICRA 2023
Subtitle of host publicationIEEE International Conference on Robotics and Automation
PublisherInstitute of Electrical and Electronics Engineers Inc.
Pages5631-5637
Number of pages7
ISBN (Electronic)9798350323658
DOIs
Publication statusPublished - 2023
Event2023 IEEE International Conference on Robotics and Automation, ICRA 2023 - London, United Kingdom
Duration: 29 May 20232 Jun 2023

Publication series

NameProceedings - IEEE International Conference on Robotics and Automation
Volume2023-May
ISSN (Print)1050-4729

Conference

Conference2023 IEEE International Conference on Robotics and Automation, ICRA 2023
Country/TerritoryUnited Kingdom
CityLondon
Period29/05/232/06/23

Bibliographical note

Publisher Copyright:
© 2023 IEEE.

Funding

ACKNOWLEDGEMENT This work is supported by Istanbul Technical University BAP Grant NO: MOA-2019-42321. We gratefully thank Eatron Technologies for their technical support.

FundersFunder number
Istanbul Teknik ÜniversitesiMOA-2019-42321

    Keywords

    • Autonomous Driving
    • Black-Box Verification
    • Deep Reinforcement Learning

    Fingerprint

    Dive into the research topics of 'Self-Improving Safety Performance of Reinforcement Learning Based Driving with Black-Box Verification Algorithms'. Together they form a unique fingerprint.

    Cite this