A twin delayed deep deterministic-based control method for a full vehicle semi-active suspension system
- 1Henan College of Transportation, Zhengzhou, China
- 2China Construction Fifth Bureau (HENAN) Construction Co., Ltd., Zhengzhou, China
- 3School of Civil Engineering, Zhengzhou University, Zhengzhou, China
Abstract
To improve the convergence issues of deep reinforcement learning in the control of vehicle semi-active suspension, and enhance the control effectiveness of the vehicle’s semi-active suspension, a Twin Delayed Deep Deterministic-based control framework tailored for the high-dimensional nonlinear dynamics of the vehicle semi-active suspension model is proposed. Rather than merely applying standard deep reinforcement learning, this study introduces a domain-specific state-action mapping mechanism and a multi-objective reward function to physically constrain the neural network outputs and coordinate the full-vehicle dynamic responses. The simulation program for 7-degree-of-freedom vehicle semi-active suspension model embedded in the proposed control method is developed using MATLAB software. The validity of the proposed control method is verified using the developed program under different road roughness and speeds. The results indicate that compared to the passive suspension the classical linear Skyhook control method and the Fuzzy control method the proposed Twin Delayed Deep Deterministic based method achieves superior control effectiveness. Specifically in terms of the vehicle body vertical acceleration the proposed method achieves a reduction of 49.41% compared to the passive suspension and 9.33% compared to the Skyhook method on a Class A road at 60 km/h a reduction of 24.82% compared to the passive suspension on a Class A road at 120 km/h and a reduction of 3.19% compared to the Skyhook method on a Class B road at 60 km/h. Furthermore compared to the traditional Fuzzy control method the proposed method demonstrates superior vibration suppression effectiveness across all evaluated operating conditions achieving significantly lower frequency weighted vertical root mean square accelerations. Furthermore the power spectral density analysis demonstrates that the proposed method significantly suppresses the low frequency resonance of the sprung mass across varying road conditions compared to the passive Skyhook and Fuzzy methods, further validating its superior broadband vibration isolation capability. This demonstrates that the Twin Delayed Deep Deterministic-based control method for the vehicle semi-active suspension model has the advantages of fast convergence speed and good control effectiveness under road surface excitation.
Related
- Trajectory tracking control of unmanned aerial vehicle with disturbance observer and robust adaptive neural dynamic surface
- Genetic control couples TGFβ-independent pluripotent stem cell renewal to deterministic forward programming across germ layers
- Predictors of Delayed Cleft Lip and Palate Surgery: A Case-Control Study in Addis Ababa, Ethiopia.
- Advantages and Limitations of Yellow Sticky Traps as a Pest Control Method in Pollinator-Friendly Rapeseed Agroecosystems: A Pilot Study in Northern Italy
- Suspension culture drives metabolic rewiring in human pluripotent stem cells
BibTeX
@article{wan2026twin,
title = {{A twin delayed deep deterministic-based control method for a full vehicle semi-active suspension system}},
author = {Yongjun Wan and Xiaoming Wang and Gang Zhi and Lei Yang and Juye Wang and Xiangcheng Zhang and Yan Zhang},
journal = {PLOS One},
year = {2026},
month = sep,
doi = {10.1371/journal.pone.0359102},
url = {https://doi.org/10.1371/journal.pone.0359102},
note = {CC BY 4.0},
}
RIS
TY - JOUR
TI - A twin delayed deep deterministic-based control method for a full vehicle semi-active suspension system
AU - Yongjun Wan
AU - Xiaoming Wang
AU - Gang Zhi
AU - Lei Yang
AU - Juye Wang
AU - Xiangcheng Zhang
AU - Yan Zhang
PY - 2026
DA - 2026/09/25
JO - PLOS One
DO - 10.1371/journal.pone.0359102
UR - https://doi.org/10.1371/journal.pone.0359102
AB - To improve the convergence issues of deep reinforcement learning in the control of vehicle semi-active suspension, and enhance the control effectiveness of the vehicle’s semi-active suspension, a Twin Delayed Deep Deterministic-based control framework tailored for the high-dimensional nonlinear dynamics of the vehicle semi-active suspension model is proposed. Rather than merely applying standard deep reinforcement learning, this study introduces a domain-specific state-action mapping mechanism and a multi-objective reward function to physically constrain the neural network outputs and coordinate the full-vehicle dynamic responses. The simulation program for 7-degree-of-freedom vehicle semi-active suspension model embedded in the proposed control method is developed using MATLAB software. The validity of the proposed control method is verified using the developed program under different road roughness and speeds. The results indicate that compared to the passive suspension the classical linear Skyhook control method and the Fuzzy control method the proposed Twin Delayed Deep Deterministic based method achieves superior control effectiveness. Specifically in terms of the vehicle body vertical acceleration the proposed method achieves a reduction of 49.41% compared to the passive suspension and 9.33% compared to the Skyhook method on a Class A road at 60 km/h a reduction of 24.82% compared to the passive suspension on a Class A road at 120 km/h and a reduction of 3.19% compared to the Skyhook method on a Class B road at 60 km/h. Furthermore compared to the traditional Fuzzy control method the proposed method demonstrates superior vibration suppression effectiveness across all evaluated operating conditions achieving significantly lower frequency weighted vertical root mean square accelerations. Furthermore the power spectral density analysis demonstrates that the proposed method significantly suppresses the low frequency resonance of the sprung mass across varying road conditions compared to the passive Skyhook and Fuzzy methods, further validating its superior broadband vibration isolation capability. This demonstrates that the Twin Delayed Deep Deterministic-based control method for the vehicle semi-active suspension model has the advantages of fast convergence speed and good control effectiveness under road surface excitation.
SN - 1932-6203
ER -