In January 22 and 23, 2026, the Conference of Rectors of Universities in South-West Europe (Conferencia de Rectores de las Universidades del Suroeste Europeo – CRUSOE), partnering with SUMLAB research group, organised the 1st International Congress of Mobility and Intelligent Transport (Sustainable Mobility Forum 2026), held in the University of Cantabria in Santander, Spain. This event aimed to explore the cutting edge of intelligent transport systems and sustainable mobility solutions, bringing together researchers, professionals, public institutions, and industry leaders.

Àngel Ruiz, a member of CoIoTIA team, attended the conference to present a use case developed in the context of the CoIoTIA project (PID2022-141813OB-I00). This work faced the last-mile routing problem by dividing the delivery region in different geographical areas with their own Deep Learning (DL) model associated to improve the route generated by a single general model. To test this approach, an encoder-decoder architecture has been developed, with a Graph Neural Network (GNN) acting as the encoder and a Pointer Network (PN) working as a decoder. The conclusion is that the zone-based approach outperforms the single-model option, drastically reducing the errors in the ground-truth route predictions.

The paper will be soon available in the Conference Proceedings. However, a pre-print is already publicly available in arXiv.

Ruiz-Fas, À., Granell, C., Ramos, J. F., Huerta, J., & Trilles, S. (2026). A zone-based training approach for last-mile routing using Graph Neural Networks and Pointer Networks. arXiv. https://arxiv.org/abs/2601.04705