Towards Developing Gripper to obtain Dexterous Manipulation

Artificial hands or grippers are essential elements in many robotic systems, such as, humanoid, industry, social robot, space robot, mobile robot, surgery and so on. As humans, we use our hands in different ways and can perform various maneuvers such as writing, altering posture of an object in-hand without having difficulties. Most of our daily activities are dependent on the prehensile and non-prehensile capabilities of our hand. Therefore, the human hand is the central motivation of grasping and manipulation, and has been explicitly studied from many perspectives such as, from the design of complex actuation, synergy, use of soft material, sensors, etc; however to obtain the adaptability to a plurality of objects along with the capabilities of in-hand manipulation of our hand in a grasping device is not easy, and not fully evaluated by any developed gripper. Industrial researchers primarily use rigid materials and heavy actuators in the design for repeatability, reliability to meet dexterity, precision, time requirements where the required flexibility to manipulate object in-hand is typically absent. On the other hand, anthropomorphic hands are generally developed by soft materials. However they are not deployed for manipulation mainly due to the presence of numerous sensors and consequent control complexity of under-actuated mechanisms that significantly reduce speed and time requirements of industrial demand. Hence, developing artificial hands or grippers with prehensile capabilities and dexterity similar to human like hands is challenging, and it urges combined contributions from multiple disciplines such as, kinematics, dynamics, control, machine learning and so on. Therefore, capabilities of artificial hands in general have been constrained to some specific tasks according to their target applications, such as grasping (in biomimetic hands) or speed/precision in a pick and place (in industrial grippers). Robotic grippers developed during last decades are mostly aimed to solve grasping complexities of several objects as their primary objective. However, due to the increasing demands of industries, many issues are rising and remain unsolved such as in-hand manipulation and placing object with appropriate posture. Operations like twisting, altering orientation of object within-hand, require significant dexterity of the gripper that must be achieved from a compact mechanical design at the first place. Along with manipulation, speed is also required in many robotic applications. Therefore, for the available speed and design simplicity, nonprehensile or dynamic manipulation is widely exploited. The nonprehensile approach however, does not focus on stable grasping in general. Also, nonprehensile or dynamic manipulation often exceeds robot’s kinematic workspace, which additionally urges installation of high speed feedback and robust control. Hence, these approaches are inapplicable especially when, the requirements are grasp oriented such as, precise posture change of a payload in-hand, placing payload afterward according to a strict final configuration. Also, addressing critical payload such as egg, contacts (between gripper and egg) cannot be broken completely during manipulation. Moreover, theoretical analysis, such as contact kinematics, grasp stability cannot predict the nonholonomic behaviors, and therefore, uncertainties are always present to restrict a maneuver, even though the gripper is capable of doing the task. From a technical point of view, in-hand manipulation or within-hand dexterity of a gripper significantly isolates grasping and manipulation skills from the dependencies on contact type, a priory knowledge of object model, configurations such as initial or final postures and also additional environmental constraints like disturbance, that may causes breaking of contacts between object and finger. Hence, the property (in-hand manipulation) is important for a gripper in order to obtain human hand skill. In this research, these problems (to obtain speed, flexibility to a plurality of grasps, within-hand dexterity in a single gripper) have been tackled in a novel way. A gripper platform named Dexclar (DEXterous reConfigurable moduLAR) has been developed in order to study in-hand manipulation, and a generic spherical payload has been considered at the first place. Dexclar is mechanism-centric and it exploits modularity and reconfigurability to the aim of achieving within-hand dexterity rather than utilizing soft materials. And hence, precision, speed are also achievable from the platform. The platform can perform several grasps (pinching, form closure, force closure) and address a very important issue of releasing payload with final posture/ configuration after manipulation. By exploiting 16 degrees of freedom (DoF), Dexclar is capable to provide 6 DoF motions to a generic spherical or ellipsoidal payload. And since a mechanism is reliable, repeatable once it has been properly synthesized, precision and speed are also obtainable from them. Hence Dexclar is an ideal starting point to study within-hand dexterity from kinematic point of view. As the final aim is to develop specific grippers (having the above capabilities) by exploiting Dexclar, a highly dexterous but simply constructed reconfigurable platform named VARO-fi (VARiable Orientable fingers with translation) is proposed, which can be used as an industrial end-effector, as well as an alternative of bio-inspired gripper in many robotic applications. The robust four fingered VARO-fi addresses grasp, in-hand manipulation and release (payload with desired configuration) of plurality of payloads, as demonstrated in this thesis. Last but not the least, several tools and end-effectors have been constructed to study prehensile and non-prehensile manipulation, thanks to Bayer Robotic challenge 2017, where the feasibility and their potentiality to use them in an industrial environment have been validated. The above mentioned research will enhance a new dimension for designing grippers with the properties of dexterity and flexibility at the same time, without explicit theoretical analysis, algorithms, as those are difficult to implement and sometime not feasible for real systems