• dual arm manipulator
• high-level planning
• intra-logistic
• logistic
• object manipulation
• path planning

In the last decade the requests of a transition to automation that can be integrated with current warehouse facilities for intra-logistic processes are steadily increasing. These services must respond quickly to changing market needs, be scalable, flexible, reliable and safe yet efficient in environments shared with humans.
This thesis addresses the problem of picking objects from single-product pallets and building multi-product pallets, in order to fill the gap between the mass distribution and the little distribution. The robot employed is a dual arm manipulator equipped with a robotic underactuated hand for manipulation tasks and a minimal flat end-effector provided with a conveyor belt to use as a support. The two manipulators are two KUKA LWR arms, each with 7 degrees of freedom.
The objective is the design and implementation of a high level planner to evaluate a manipulation policy for the product handling based on pre-defined elementary manipulation skills. The high level planner builds a planning graph, which represents all the allowed configuration of the object and the manipulation actions that can be performed, starting by the pre-specified set of correspondences between end-effectors, objects and grasps stored in a database. Given the initial and final position of the good, the policy is found searching for the shortest path from the initial to the final state, using the Dijkstra’s algorithm. The sequence of the selected actions is then converted to a list of Cartesian waypoints that, eventually, are processed by a low level controller, which returns the robot joint trajectories.
The robot is designed to perform prehensile grasps and non-prehensile pushing actions to slide and topple objects placed on the pallet surface. The latter actions are particularly effective against objects of big dimensions that cannot be grasped single handedly, such as those present in warehouse facilities. The feasibility is ensured by corrective actions and backtracking planning in case of failure of the performed task or in presence of obstacles. The validity of the proposed planner is shown via experiments both for picking and placing tasks.