Autonomous Planning Strategies for Robotic Grasping and Manipulation

The first part of this thesis work extends a dual arm manipulation planner, developed by the University of Pisa in the context of the EU financed Project SOMA, to include environment exploiting skills. This is achieved by considering environmental constraints no more as obstacles but rather as opportunities that need to be exploited for achieving robust and stable object manipulation. New skills like sliding, tilting and edge grasping were added, remaining consistent to the pre-existing structure and formalism of the framework.Experimental evaluation of the extended planning system, performed through testing on the Vito robot (bi-manual robotic platform of the Research Center E. Piaggio), clearly displayed improved range of applicability, efficiency and robustness of the planner.Formal and experimental comparison of the extended version of the UNIPI Planner with a similar framework, developed by the Technical University of Berlin, was carried out. The comparison also lead to the integration work of the TUB planner on to the UNIPI robotic setup. Through testing, both planning systems were found to be innovative and computationally efficient in their different, yet compliant, goals of object manipulation and grasping respectively. The exploitation of the environmental constraints was found to aid considerably in both planning algorithms. Secondly, the focus is turned mainly towards robotic grasping: it is known that traditional robotic hand grasping techniques for achieving an enveloping grasp require the application of some pressure with the palm before closing the hand, so that the object is not moved away by the fingers. However, this method is not really the best for soft objects as it could potentially damage the manipulandum. So, the second part of the thesis consists in the implementation of a novel type of soft grasping strategy, the Adaptive Grasping, which is a new human inspired grasping technique that has not been used in robotics before. Following the results of neurological studies which show that human grasping is never an action performed solely by the hand, but a combined action by the arm and the hand coordinated by the neuromotor system, the Adaptive Grasping tries to go beyond the simple hand closing motion. We get a simple tactile feedback by using acceleration measurements of the fingers; then, according to the touching finger the robot can choose or modify the grasping policy and by making use of a synced motion of both the hand and the arm an efficient and robust grasp, which adjusts the hand to the object, is achieved. The efficiency of this type of grasping is confermed by experimental results of several robotic platforms