Utilisation d'algorithmes d'intelligence artificielle pour l'identification et la localisation de colis et leur saisie par un torse humanoïde dans des opérations de dépalettisation

In the context of Logistics 4.0, the robotisation of handling operations is developing but still needs to resolve specific scientific issues to succeed in replacing the dexterity of human operators. The use of a humanoid torso has proved to be a relevant solution for picking up parcels of varying sizes. However, the complete robotisation of depalletising operations by the humanoid torso remains a complex problem when the boxes are set up in layers and juxtaposed. First of all, the boxes have to be identified and located unit by unit so that they can then be handled. The work presented in this paper shows that the image analysis approach is reaching its limits and introduces the use of artificial intelligence and deep learning on the sensor data provided by an RGD-D camera. The design and implementation of this method are detailed and the experimental results are presented.