Autonomous Container Unloading for Heavy and Voluminous Objects in Decathlon Logistics Fulfillment Centers

Efficient and reliable merchandise transport is the backbone of the global economy, ensuring that goods move seamlessly from production to consumer. Distribution centers are the critical nodes in this network, where complex processes such as receiving, sorting, and shipping are performed, many of which are still highly dependent on manual labor. Among these processes, the manual unloading of merchandise from inbound containers remains one of the most physically demanding and operationally critical bottlenecks. Although some robotic systems have been developed to solve the unloading task, our research on state-of-the-art applications shows that the currently available systems cannot handle loads of more than 29 kg. To address this gap, this thesis presents a robotic platform for the unloading of unpalletized containers up to 35 kg and dimensions up to 800 × 800 × 1200 mm, in the specific context of the inbound operation in Decathlon distribution centers. The robotic system presented includes a path planning procedure for the container unloading task together with a perception system that provides information about the robot surroundings. The trajectory planning for the pickand-place operation was performed using a Rapidly-exploring Random Trees Star (RRT*)-based algorithm together with a task subdivision. For the robot’s perception, a RetinaNet network architecture was trained using cardboard images for the 2D detection of the cartons and was paired with a custom projection strategy to identify and localize cartons in the 3D space. In this thesis, we present a series of extensive simulation tests, in which we analyze different unloading scenarios, including different box geometries and planning algorithms. The results obtained demonstrate that the proposed path planning procedure can be operated at an unloading rate of 140 boxes per hour for boxes with dimensions up to 600 × 600 × 750 mm which covers around 80 % of the merchandise received in Decathlon warehouses. Moreover, results show that the robotic system developed can perform the planning for boxes up to 800 × 800 × 1200 mm at a reduced rate of 68 boxes per hour, covering the totality of the boxes being unloaded in the selected workspace in Decathlon warehouses, therefore, removing completely the ergonomic risk for Decathlon operators. The developed perception system shows a precision and recall of 97 and 93 %, respectively for the detection of boxes from a videocamera input. The simulation results obtained show the validity of the proposed approach for the unloading of heavy and voluminous packages using a robotic platform in Decathlon warehouses. Moreover, the modular architecture of the robotic system lays the foundations for a general-purpose container unloading system, allowing to easily integrate newly developed functionalities and customizations.