• cnn
• control
• deep learning
• manipulation
• neural network
• planning
• robotics
• vision

Although production processes have reached a high level of automation, logistics still requires manual operations and hand-operated forklifts that can cause inefficiencies. The challenge that prevents a full logistic automation is given by the high variety of products stored in warehouses, the handling thereof requires extremely flexible automation solutions. This thesis is included in the ILIAD project, which has the design of an autonomous system for de-palletizing and palletizing of heterogeneous goods as one of the main purposes. The solution designed for the project is a dual-arm robotic platform provided with a Pisa/IIT SoftHand and a Velvet Tray. The design of the robot structure and the definition of a set of base functions for the manipulation system have been inspired by the techniques the human operators adopt in Pick-and-Place (P&P) operations. Experiments carried out have shown that the robot is capable to efficiently perform the picking operations in a structured environment scenario, where the type and the position of the objects to manipulate are known and sent to the robot. However, modern robotic solutions have to be robust and effective in unstructured environments and need a perception system. In order to improve robot reliability and achieve a fully autonomous solution, suitable feedback strategies based, e.g., on vision and force feedback have to be implemented.