• Model Predictive Control
• Non-Prehensile Manipulation
• optimization

The goal of this work is to evaluate the performances of online numerical optimal control techniques for robotic control applications, specifically for nonprehensile manipulation tasks.
Nonprehensile manipulation is the process of manipulating a part without a form- or force-closure grasp. Without such a grasp, the part is free to roll, slide, or break contact with the robot(s) manipulating it.
Among the realm of nonprehensile manipulation this work focus on fully dynamical task which include impulsive contact and ballistic phase.
First the main difficulties and issues to overcome are listed.
Then available solutions both in literature both biological system like the human motion control system are analyzed and imitated.
Finally some implementation are tested in benchmark cases with descending complexity. Some considerations are expressed regarding the approach to this kind of under-actuated hybrid problems in a systematic way.
Open loop optimization have been tested on a 2D non-prehensile system composed by a capsule and a plane.
Closed loop online techniques have been successfully applied in simulation to a simplified 2D case, composed by a paddle and a ball.
For an even simpler case, 1D ball bouncing system, an experimental apparatus have been designed and build and online optimization algorithm has been tested in real world.