Article: Intelligent obstacle avoidance control method for autonomous vehicles based on improved SAC algorithm Journal: International Journal of Vehicle Design (IJVD) 2025 Vol.99 No.5 pp.1 - 19 Abstract: To improve the success rate of collision avoidance for autonomous vehicles and shorten response time, an intelligent obstacle avoidance control method based on an improved SAC algorithm is proposed. This method is based on a self-organising cluster model, integrating short-range repulsion, medium-range velocity calibration, and obstacle avoidance rules to achieve collision-free cluster collaboration. The conventional SAC algorithm adopts the AC framework to maximise the expected reward and entropy value while reducing the estimation bias of the value function through the value network component. On this basis, the PER-SAC method is proposed, which integrates priority experience replay (PER) and importance sampling weight strategy while optimising network structure, reward and punishment functions, continuous state and action space design. Additionally, transfer learning is incorporated. The experimental results demonstrate the effectiveness of this method, achieving a collision avoidance success rate of 97%, with a maximum response time of just 0.54 s. Inderscience Publishers - linking academia, business and industry through research

Title: Intelligent obstacle avoidance control method for autonomous vehicles based on improved SAC algorithm

Authors: Yulin Ma; Yide Qian; Teng Ma; Yicheng Li; Jian Wan

Addresses: School of Mechanical and Automotive Engineering, Anhui Polytechnic University, Wuhu, 241000, China ' School of Mechanical and Automotive Engineering, Anhui Polytechnic University, Wuhu, 241000, China ' School of Mechanical and Automotive Engineering, Anhui Polytechnic University, Wuhu, 241000, China ' Automotive Engineering Research Institute, Jiangsu University, Zhenjiang, 212013, China ' School of Software Engineering, Jinling Institute of Technology, Jiangsu Key Laboratory of Digital Intelligent Low-Carbon Transportation, Nanjing, 211169, China

Abstract: To improve the success rate of collision avoidance for autonomous vehicles and shorten response time, an intelligent obstacle avoidance control method based on an improved SAC algorithm is proposed. This method is based on a self-organising cluster model, integrating short-range repulsion, medium-range velocity calibration, and obstacle avoidance rules to achieve collision-free cluster collaboration. The conventional SAC algorithm adopts the AC framework to maximise the expected reward and entropy value while reducing the estimation bias of the value function through the value network component. On this basis, the PER-SAC method is proposed, which integrates priority experience replay (PER) and importance sampling weight strategy while optimising network structure, reward and punishment functions, continuous state and action space design. Additionally, transfer learning is incorporated. The experimental results demonstrate the effectiveness of this method, achieving a collision avoidance success rate of 97%, with a maximum response time of just 0.54 s.

Keywords: improved SAC algorithm; autonomous vehicles; intelligent obstacle avoidance control; PER; priority experience replay.

DOI: 10.1504/IJVD.2025.151524

International Journal of Vehicle Design, 2025 Vol.99 No.5, pp.1 - 19

Received: 25 Dec 2024
Accepted: 15 Aug 2025
Published online: 04 Feb 2026 *

Title: Intelligent obstacle avoidance control method for autonomous vehicles based on improved SAC algorithm

Keep up-to-date