Behavioral decision-making for autonomous driving using Soft Actor-Critic algorithm

In recent years, the advancement of artificial intelligence has significantly influenced the enhancement of automatic driving technology. Decision-making in autonomous vehicle driving behavior, based on reinforcement learning algorithms, often encounters challenges such as low sampling efficiency, p...

Full description

Saved in:
Bibliographic Details
Published in2024 36th Chinese Control and Decision Conference (CCDC) pp. 354 - 360
Main Authors Guo, Jun, Zhu, Xuefeng, Zeng, Qingrong
Format Conference Proceeding
LanguageEnglish
Published IEEE 25.05.2024
Subjects
Artificial intelligence
Autonomous driving
Autonomous vehicles
behavioral decision-making
Decision making
Navigation
Reinforcement learning
SAC
Stability analysis
Testing
Online AccessGet full text

Cover

More Information
Summary:In recent years, the advancement of artificial intelligence has significantly influenced the enhancement of automatic driving technology. Decision-making in autonomous vehicle driving behavior, based on reinforcement learning algorithms, often encounters challenges such as low sampling efficiency, prolonged learning times, and suboptimal driving stability. This paper introduces a model for automatic driving decision-making utilizing a Soft-Actor-Critic (SAC) based reinforcement learning algorithm, with the goal of achieving safe and autonomous vehicle navigation. The proposed Actor Refinement Network seeks to enhance driving stability within the model. Experimental validation of the SAC-based algorithm was performed on the Torcs autonomous driving simulation platform. Results indicate robust performance and notable generalization ability of the SAC algorithm model in the domain of autonomous driving behavioral decision-making. To further bolster vehicle stability, the output of the Actor network underwent refinement and functional verification on the simulation platform.
ISSN:1948-9447
DOI:10.1109/CCDC62350.2024.10588017