Within distribution centers, separating and sorting parcels are essential activities for managing an efficient order flow. However, these tasks are largely manual, repetitive, and potentially hazardous for operators, making their automation a strategic priority for the sector’s development.

The award-winning paper, Optimizing Parcels Sorting Through Reinforcement Learning for Intralogistics, presents a modular sorting system controlled by an artificial intelligence algorithm. The module consists of several active elements, each capable of moving parcels forward and diverting them laterally. By acting in a coordinated manner, these elements create the necessary spacing between parcels to enable the subsequent processing stages.
 

From simulation to practical application

Unlike traditional systems based on predefined rules, the movement control is managed through a reinforcement learning algorithm that autonomously learns which actions yield the best results.

Training is carried out first in a simulation environment designed to realistically reproduce parcel behavior. In this phase, the algorithm tests various strategies and receives positive or negative feedback based on its ability to achieve proper parcel separation at the output, progressively converging toward an effective behavior.

For real-world operation, the system relies on a computer vision setup composed of multiple cameras that detect the position and size of parcels in transit. The information extracted from the images is provided to the algorithm, which continuously updates its commands to the sorting module, enabling dynamic and adaptive control.

Once training is complete, the control strategy developed in simulation is transferred to a real installation and tested under operational conditions. The results show a high level of accuracy:

• About 96.5% accuracy for sorting standard parcels, similar to those used during simulation

• About 94% in more complex scenarios involving objects not present during training, such as small envelopes with different dynamic behaviors, while maintaining an operating speed of approximately 6,000 parcels per hour

This work demonstrates how advanced research can make a concrete contribution to the evolution of the logistics sector, offering automation solutions capable of increasing efficiency, safety, and reliability along the entire distribution chain.