Deep reinforcement learning (DRL) has been demonstrated to be effective for the control of mixed-autonomy platoons. However, incorporating safety considerations into DRL typically relies on designing collision avoidance-based reward functions, which lack collision-free guarantees. To address this research gap, we introduce a differentiable safety layer to convert DRL actions to safe actions with collision-free guarantees, which relies on a differentiable quadratic programming problem featuring Control Barrier Function-based safety constraints. The proposed safety-informed DRL approach explicitly integrates safety constraints into the training process and enables our method to adapt to varying safety-critical scenarios. Simulation results demonstrate that our proposed method effectively ensures CAV safety in mixed platoon environments while simultaneously enhancing traffic capacity and string stability. Furthermore, the safety layer expedites training by reducing the range of exploration.


    Access

    Check access

    Check availability in my library

    Order at Subito €


    Export, share and cite



    Title :

    Safe Reinforcement Learning for Mixed-Autonomy Platoon Control


    Contributors:
    Zhou, Jingyuan (author) / Yan, Longhao (author) / Yang, Kaidi (author)


    Publication date :

    2023-09-24


    Size :

    605761 byte





    Type of media :

    Conference paper


    Type of material :

    Electronic Resource


    Language :

    English



    Reinforcement Learning for Mixed Autonomy Intersections

    Yan, Zhongxia / Wu, Cathy | IEEE | 2021


    AVDDPG: Federated reinforcement learning applied to autonomous platoon control

    Boin, Christian / Lei, Lei / Yang, Simon X. | ArXiv | 2022

    Free access



    Safe platoon control of automated highway systems

    Huang, A.C. / Chen, Y.J. | Tema Archive | 2001