A novel technique for effective grasp synthesis and the introduction of a computational toolkit for exploring the grasping problem
In today's industry, numerous types of industrial grippers are used to carry out grasping tasks that demand a great deal precision and reliability. Thus, the majority of commercial grippers can meet industry demands as long as the work environment is highly structured and controlled. Nevertheless, their capabilities become insufficient in the presence of uncertainties and so the quality of the products is significantly degraded. On the contrary, the flexibility of dexterous robotic hand devices may provide the necessary means for dealing with uncertain and highly complex work environments. But, the inherent redundancy of highly-articulated robotic hands makes it very challenging to synthesize a suitable grasp that meets the desired task requirements. This problem has motivated the development of various grasp synthesis methods for determining grasps that realize a hand's full potential; although this is still a work in progress.
In addition, the use of physical robotic prototypes for evaluating grasp synthesis methods is not always feasible because they are scarce and present numerous technical and financial challenges. Thus, preliminary studies of robotic grasping methods would be greatly benefited from simulation software since it offers an easier route for investigating the effectiveness of several grasping methods. Also, a software alternative is flexible enough to accommodate large sets of test scenarios that can lead to stronger conclusions about a particular technique; this is not the case for experimental test-bed.
The work in this thesis addresses two main topics with the aim of presenting suitable solutions to the problems described above. The first goal of this thesis is to introduce a new technique for grasp synthesis that is applicable to dexterous hand models of any kinematic configuration and geometry. This technique relies heavily on continuous collision detection (CCD) in order to find suitable fingers postures candidates that may increase the quality of the grasps. In addition, it utilizes a search procedure involving 6-dimensional hyper-plane half-spaces for seeking and selecting finger posture candidates that increase the grasp quality. Also, grasps are rejected whenever interferences between links of the hand are detected. The effectiveness of this method is demonstrated through three experiments involving three different hand models undergoing similar task conditions.
The second goal in this thesis is to present and describe the ongoing development of a computational toolkit that incorporates a sufficient number of tools for examining and simulating a multitude of grasping tasks. Moreover, the addition of several modules allow conducting computations related to computational geometry, grasp mechanics, collision detection and robust physics simulation; which aid in the development of grasp synthesis methods. This toolkit also relies on scripting in order to describe an entire task, with various goals and constraints, in terms of compatible functions or word commands written on a script. By editing the script, it's possible to specify a new set of goals and constraint very quickly and the results become available almost immediately. In the end, an experiment that combines the grasp synthesis technique with the physics simulator serves to demonstrate the benefits of using the tools in the computational library towards the investigation of robotic grasping.