Scholarlink

<p>This work focuses on the intersection coordination control problem for an autonomous heterogeneous fleet in open pit mines. We developed a reinforcement learning method called multiagent actor-critic proximal policy optimization (AC-PPO) based on clipped proximal policy optimization (clipped-PPO), with the objective of minimizing the finish time for the fleet to pass through the intersection area. The proposed approach addresses the problem of controlling the operation of a heterogeneous truck fleet by calculating the priorities of all trucks in the intersection area and adding them to the state space of the trucks. Additionally, the selected state space can be utilized for intersections. Furthermore, we have enhanced the parameter updating method of clipped-PPO by creating two separate networks to update the parameters associated with the policy network and the value function network. The simulation experiments demonstrate that our training model can adapt to various scenarios. Additionally, AC-PPO outperforms clipped-PPO while utilizing fewer network parameters.</p>

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

Coordination Scheme for Autonomous Heterogeneous Mining Fleets at Open Pit Mine Intersections by Proximal Policy Optimization

  • 期刊论文

  • 发表日期：

    2025年4月4日

查看全文
  • Zhichao Wang,
  • Jue Yang,
  • Lei Liu

摘要

This work focuses on the intersection coordination control problem for an autonomous heterogeneous fleet in open pit mines. We developed a reinforcement learning method called multiagent actor-critic proximal policy optimization (AC-PPO) based on clipped proximal policy optimization (clipped-PPO), with the objective of minimizing the finish time for the fleet to pass through the intersection area. The proposed approach addresses the problem of controlling the operation of a heterogeneous truck fleet by calculating the priorities of all trucks in the intersection area and adding them to the state space of the trucks. Additionally, the selected state space can be utilized for intersections. Furthermore, we have enhanced the parameter updating method of clipped-PPO by creating two separate networks to update the parameters associated with the policy network and the value function network. The simulation experiments demonstrate that our training model can adapt to various scenarios. Additionally, AC-PPO outperforms clipped-PPO while utilizing fewer network parameters.