.png){kind=icon}
.png){kind=icon}
Return
Event-Triggered MPC for Nonlinear Connected Autonomous Vehicle Platoons With a Deep Reinforcement Learning Approach
.png){kind=icon}
PRE
AI
X
S
J
X
关
DOI：10.1109/tvt.2026.3665584.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
This paper proposes a reinforcement learning-enhanced event-triggered model predictive control (EMPC) framework for nonlinear networked vehicular platoon systems. The developed architecture integrates deep reinforcement learning (DRL) with distributed EMPC (DEMPC) to autonomously optimize event-triggered thresholds, thereby reducing dependence on prior knowledge of system dynamics and communication constraints. To enhance policy exploration and training efficiency, the framework incorporates priority experience replay (PER) buffers and long short-term memory (LSTM) networks for temporal feature extraction. A nonlinear DEMPC scheme is formulated with distributed controllers constrained by neighbor state interactions governed by the vehicular communication topology. Furthermore, closed-loop stability under unidirectional topology is rigorously established via Lyapunov analysis based on the constructed cost function. Three prominent DRL algorithms-Double Deep Q-Network (DDQN), Advantage Actor-Critic (A2C), and Soft Actor-Critic (SAC)-are systematically implemented and compared within this control paradigm. Both simulation and experimental results demonstrate that all RL-enhanced EMPC controllers successfully achieve accurate path tracking in nonlinear multi-vehicle scenarios. The final results showed that the triggering frequency of the DDQN, A2C, and SAC algorithms decreased by 33.91%, 21.57%, and 35.16%, respectively, compared to the baseline threshold method.
Keywords:
Event-triggered MPC
platoon control
deep reinforcement learning
autonomous vehicles
Journal
IF:
7.1
Papers:
1.7W
Citations:
6.6W
Organization
Cited Papers