AGV path planning and optimization with deep reinforcement learning model

The rapid progress in global port construction has resulted in a significant rise in container throughput. As the development of automated container terminals continues, it is imperative to enhance the intelligence and environmental friendliness of Automated Guided Vehicles (AGVs). This paper focuses on addressing the AGV path planning problem by employing deep reinforcement learning. The primary objectives are to optimize data collection efficiency and minimize battery consumption. The diverse layouts of terminal environments are represented using RGB images, and path learning is incorporated with a meticulously designed reward function. Here, we perform 1000 simulations using Monte Carlo to evaluate the parameters for these four scenarios. The evaluation showed that the terminal path planning battery consumption averaged around 40 to 56, the data collection rates were all above 80% on average, and more than 99% of them resulted in optimal solutions in a 32×32 size environment. The algorithm is able to reach and collect the data as fast as possible.