Scholarlink

As the demand of automated container terminals increases, automated guided vehicles (AGVs) scheduling in twin-lift operations unloading tasks becomes complex due to multi-vehicle collaboration, dynamic task allocation, and time window constraints. Traditional static optimization methods are difficult to cope with the real-time changing environment. In this paper, we propose a deep reinforcement learning-based the Deep Deterministic Policy Gradient (DDPG) algorithm to optimize the AGVs scheduling problem in the twin-lift operations unloading task by constructing a multi-stage Markov decision process (MDP) model with task allocation, vehicle pair scheduling and time window constraints. Compared to mixed integer programming and heuristic algorithms, DDPG achieves a 100% task completion rate in small to medium-scale scenarios, shortens the total operation time to 1800.50 s, minimizes synchronization error to 12.30 s, and lowers the usage of outsourced vehicles by 28.6%. These results confirm that DDPG is a highly effective optimization strategy for intelligent scheduling in automated container terminals.

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

Deep Reinforcement Learning-Based AGVs Scheduling Optimization for Twin-Lift Operations Unloading Tasks in Automated Container Terminals

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Jingze Li,
  • Zeyu Zhang,
  • Ai Mu,
  • Xiaofan Lai

摘要

As the demand of automated container terminals increases, automated guided vehicles (AGVs) scheduling in twin-lift operations unloading tasks becomes complex due to multi-vehicle collaboration, dynamic task allocation, and time window constraints. Traditional static optimization methods are difficult to cope with the real-time changing environment. In this paper, we propose a deep reinforcement learning-based the Deep Deterministic Policy Gradient (DDPG) algorithm to optimize the AGVs scheduling problem in the twin-lift operations unloading task by constructing a multi-stage Markov decision process (MDP) model with task allocation, vehicle pair scheduling and time window constraints. Compared to mixed integer programming and heuristic algorithms, DDPG achieves a 100% task completion rate in small to medium-scale scenarios, shortens the total operation time to 1800.50 s, minimizes synchronization error to 12.30 s, and lowers the usage of outsourced vehicles by 28.6%. These results confirm that DDPG is a highly effective optimization strategy for intelligent scheduling in automated container terminals.