Adding flexibility to the vehicle routing problem formulation

What is it about?

Reinforcement Learning (RL) is an artificial intelligence technique widely used in areas such as games or robotics, as it can learn from various scenarios and use past experiences to handle previously unseen circumstances. This study aims to apply its benefits to the Vehicle Routing Problem (VRP) while mitigating some of its drawbacks, such as the high time needed to train this kind of artificial intelligence technique. To solve this, the study proposes to add a reusability factor to the trained models, so the same model can be used in a different use case than the one it was originally trained in.

Featured Image

Photo by CHUTTERSNAP on Unsplash

Photo by CHUTTERSNAP on Unsplash

Why is it important?

VRP is one of the most universally studied combinatorial problems and many techniques have proven to be very competent at solving it. However, most of these techniques do have the drawback of needing to be adapted depending on the specifics of the problem. For this reason, due to the benefits of RL, we wanted to study the viability of the technique to solve vehicle routing. RL and VRP are two fields that have seen little interaction with one another, and in spite of having found success in specific use cases, no general-purpose technique has been proposed yet. Through this study, we have created a simple RL environment able to solve simple cases, and then carry that knowledge to completely different ones. Although no breaking results have been made, thanks to the conducted research, we have observed RL to be suitable for solving VRP, possessing even the potential to outperform current popular techniques.

Perspectives