Factories and manufacturing centers rely on automated guided vehicles (AGVs) to move materials from one place to another with limited human oversight. These vehicles are a valuable asset for increasing productivity while freeing up workers to focus on other tasks. AGVs that are not coordinated efficiently, however, can cause slowed manufacturing, under-utilization of resources, and bottlenecks all along a production line. Using a manufacturing center simulation, we demonstrate how reinforcement learning is the ideal optimization solution for AGV fleets.


The featured manufacturing center uses AGVs to move component parts to machines. Arriving components need to be taken to the correct location depending on the manufacturing center’s processing sequence. The fleet needed to be optimized so that it increased the rate at which parts were taken to the right place and maximized the output of finished products.

Why Pathmind Was Needed

Reinforcement learning is the best optimization solution for AGV fleets for two main reasons: the number of agents that need to be coordinated and the production variables that need to be taken into account. Other optimization methods struggle when tasked with conducting large numbers of moving pieces that need to work well both individually and as a unit. Likewise, baseline heuristics cannot account for the dynamics of real-world production, such as maintenance shutdowns and irregular processing times.

AGV model screenshot


Coordinating the AGVs with reinforcement learning produced an efficient fleet capable of increasing throughput while better utilizing resources. This success demonstrates how AI is the best optimization solution when you need to take production variables into account and optimize multiple agents simultaneously. Compared to a shortest queue heuristic, the AI policy outperformed by nearly 78%.

Want to learn more? View this model on AnyLogic Cloud.

Then, explore the model and walk through the steps of adding reinforcement learning with our AGV Tutorial.