Scholarlink

Multi-vehicle cooperative perception significantly enhances the accuracy and coverage of environmental sensing in intelligent vehicle platoons. However, the random arrival of sensing tasks and the heterogeneity of available resources often lead to inefficient scheduling and system instability. To address these challenges, we propose a deep reinforcement learning-based task scheduling framework tailored for dynamic vehicular edge environments. First, a hierarchical scheduling architecture is established with a Head Vehicle (HV) that collects real-time system states and perception task queues. Second, an improved Proximal Policy Optimization (PPO) algorithm is employed, where the HV generates optimal scheduling actions, including task assignments, transmission power, and communication intervals-based on compressed state representations. Third, a carefully designed reward function considers both average latency and task response fluctuations, ensuring long-term system stability. Finally, the policy network is iteratively optimized through multi-round interactions to improve adaptability under dynamic workloads. Simulation results demonstrate that our approach reduces average task response delay by up to 24% compared to conventional methods, while maintaining superior robustness and efficiency.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adaptive Cooperative Perception Task Scheduling for Vehicle Platoons via PPO-Based Deep Reinforcement Learning

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Fenghui Zhang,
  • Jiaxin Ma,
  • Ao Zhou,
  • Yu Zong,
  • Qiu Xu,
  • Huaqiang Xi

摘要

Multi-vehicle cooperative perception significantly enhances the accuracy and coverage of environmental sensing in intelligent vehicle platoons. However, the random arrival of sensing tasks and the heterogeneity of available resources often lead to inefficient scheduling and system instability. To address these challenges, we propose a deep reinforcement learning-based task scheduling framework tailored for dynamic vehicular edge environments. First, a hierarchical scheduling architecture is established with a Head Vehicle (HV) that collects real-time system states and perception task queues. Second, an improved Proximal Policy Optimization (PPO) algorithm is employed, where the HV generates optimal scheduling actions, including task assignments, transmission power, and communication intervals-based on compressed state representations. Third, a carefully designed reward function considers both average latency and task response fluctuations, ensuring long-term system stability. Finally, the policy network is iteratively optimized through multi-round interactions to improve adaptability under dynamic workloads. Simulation results demonstrate that our approach reduces average task response delay by up to 24% compared to conventional methods, while maintaining superior robustness and efficiency.