Fleet planning is a tedious task. That is what we learned through our first interactions with an established client in the transportation sector. But through various brainstorm sessions, we came with a solution to this problem.

Recent reinforcement learning algorithms have proven to be very succesful at planning in an uncertain environment. We created a planning agent by using a phased approach and started by training in a gamified environment. It soon became clear this would work in real-world problems as well. By including more aspects of the planning process in the problem definition, the agent learns to assist its human co-workers and lets them focus on the most business-critical tasks with highest customer impact.