Toward Energy-Efficient Routing of Multiple AGVs with Multi-Agent Reinforcement Learning

Area of Science:

• Robotics
• Artificial Intelligence
• Operations Research

Background:

• Automated guided vehicles (AGVs) are crucial for logistics and manufacturing automation.
• Optimizing AGV scheduling and routing is complex, with energy efficiency often overlooked.
• Existing methods may not adequately address the dynamic and multi-agent nature of AGV systems.

Purpose of the Study:

• To develop a multi-agent reinforcement learning (MARL) algorithm for efficient AGV scheduling and routing.
• To minimize overall energy consumption in multi-AGV operations.
• To enhance AGV task assignment and path planning capabilities.

Main Methods:

• The study proposes a modified multi-agent deep deterministic policy gradient (MADDPG) algorithm tailored for AGV environments.
• A specialized reward function is designed to prioritize energy efficiency.
• The e-greedy exploration strategy is integrated to improve training convergence and performance.

Main Results:

• The proposed MARL algorithm effectively addresses multi-AGV task assignment and path planning challenges.
• Numerical experiments demonstrate superior energy efficiency compared to baseline methods (MADDPG, Q-Learning).
• The algorithm successfully balances exploration and exploitation for faster convergence and better performance.

Conclusions:

• The developed MARL algorithm offers a significant advancement in energy-efficient AGV operations.
• The approach provides a robust solution for complex scheduling and routing problems in dynamic environments.
• This research contributes to more sustainable and efficient automated logistics systems.