Identifying Object Grasps for Manipulation

Abstract

In the field of robot manipulation, grasping is an important research area. Identifying the different properties of objects is necessary for manipulating and grasping objects in real environments before being used in automation. This increases the efficiency of the tasks during execution. Robust grasps can be evaluated in simulation environment. But the simulated models and the real robotic systems have inconsistencies due to modeling and calibration errors. This necessitates measuring the success of the grasp on real systems, not just in simulated environment. The objective of the thesis remains to develop an intelligent system which is capable of generating and evaluating grasp quality metrics on real system. Due to the inaccessibility to real systems, the final results evaluate the success of the grasp data in the simulator and sequence them, to ease the testing on the real systems using these results. A detailed approach and implementation to achieve the feasible grasps, to obtain success rates of all the grasps and generate an optimal sequence of grasps is provided in this thesis. The evaluation has been performed on DLR Omnirob in simulation to validate the desired results for the same. The main results indicate identifying object grasps for a given object which are feasible and reachable. From these results, an optimal sequence of grasps are generated which can be tested on real systems.