Start research on Artificial Intelligence for Logistics

Operations of logistic companies often are a chess game between flexibility and efficiency. Plans must continuously be adjusted to match the reality as it starts to deviate from the original plans. Adjusting those plans is a puzzle that costs a lot of time and money. The research project 'DynaPlex: Deep Reinforcement Learning for Data-driven Logistics' aims to address this by making Artificial Intelligence accessible for the planning of logistics and transport. On July 15, a conglomerate of companies including ASML and Vanderlande, Den Hartogh, Ewals, Nexperia and Ahold signed a four-year research project in which 2,3 million euros is invested. The University of Twente is also involved in the project.

Solving problems

According to project leader Willem van Jaarsveld, there is much to be gained: "In fact, usually less than 50% of the orders are executed according to plan. The planners then spend the whole day solving problems with customer orders. They must do this almost entirely on their own, there is little knowledge transfer and software support. As a result, they often must reinvent the wheel, which leads to problems and frustration, both for the planners themselves and for the customers."

Game

Van Jaarsveld explains why AI is very suitable for finding an answer. "Think of it as playing a game. You make a move and then something unexpected can happen. In a game, this is a move made by your opponent. In a day-to-day planning, this might be a customer who changes an order, or a delivery that is delayed. Or a traffic jam on the highway that delays the shipment. Deep Reinforcement Learning (one of the forms of AI) can then help in different ways to figure out the next move in your planning."

In the project in which also TU/e researchers Yingqian Zhang (Information Systems) and Remco Dijkman (Operations Planning & Control) are collaborating, we are going to create a software package that makes it very easy to implement an AI that helps with specific business problems. Think of an AI that helps with transportation planning of Den Hartogh or Ewals; an AI that helps with warehouse planning with Vanderlande technology; or an AI that helps with maintenance planning of ASML machines. These kinds of software packages already exist for more traditional planning techniques (mathematical programming), so we are going to build them for AI-driven planning.