Multi-fingered Robotic Hand Grasp Planning

Being the executive component of interaction between the robot and the environment, multi-fingered dexterous hand is capable of stable grasping and dexterous manipulation of objects in different shapes through mimicking human hand both in structure and functionality, and aroused great interest from scholars at home and abroad in the past three decades. The grasp planning becomes complicated because of the increased degrees of freedom and diversity of grasp mode. In this thesis, in-depth research was performed on the robotic grasp planning issues, and a novel multi-fingered robotic hand driven by shape memory alloy was developed.The main innovative work of this thesis is summarized as follows:(1) Proposes a optimal planning algorithm for polygonal object grasping with frictional contactsWorkpiece can be treated as2D object when it is restricted to move on the workbench. Considering a polygonal object grasped by n fingers, a planning algorithm is proposed to calculate the initial positions satisfying force-closure condition. After defining the maximum external wrench in all directions a grasp can balance as a criterion, we analyzed the features of the primitive wrench set and both the geometrical and the physical significances of the grasping planning criterion in wrench space. Starting at the initial arrangement, an iterative algorithm is proposed to continuously improve grasp performance towards the optimal direction. It converges to the near-optimal grasp arrangement in less iteration and reduces the running time dramatically.(2) Enhances the ray-shooting approach to grasp evaluationCalculation of the intersection point on the boundary of convex hull of the contact wrench set with a ray emitting from the origin along a certain direction is required in the grasp evaluation problem. The geometry of the grasp wrench space can be read by the support mapping in the ray-shooting approach. Without linearization of the friction cones, the efficiency of the grasp evaluation is improved. In this thesis, we discuss some issues such as how to determine whether the origin is contained in the convex hull of the contact wrench set, how to treat the singular cases in the iteration, etc. In the same time, an enhanced algorithm with searching tree and short-circuiting technique is put forward to further accelerate the grasp evaluation algorithm.(3) Presents a novel grasp evaluation method and implements it to polyhedral objects grasp planningA novel grasp evaluation method is presented based on disturbance force rejection under the assumption that the normal component of each individual contact force is less than one. It overcomes the drawbacks of other grasp quality indices such as non-uniformity of the wrench space and a dependence on scale and choice of reference frame, besides, it incorporates the object geometry, can be visualized easily for3-D grasps. Next, we implement it to optimal grasp planning for polyhedral objects in3-D space and deduce a necessary condition for grasp quality improvement. Starting from an initial force-closure unit grasp configuration, an iterative algorithm is proposed to find the locally optimum contact points on the assigned faces of a polyhedral object.(4) Proposes a fast maximum stable grasping algorithm in dynamic force distributionWhen the grasped object endures time-varying external wrench, it requires proper contact forces to offset the external wrench and maintain balance. In order to calculate the maximum stable grasping force for each finger quickly, we divide the dynamic force distribution algorithm into two phases and speed up the on-line computation by shifting as much computation as possible from an on-line phase to an off-line phase. In the off-line phase, a nonsingular simplex set is obtained by the zone triangulation of the convex hull of the contact primitive wrench set, and the adjacent simplexes of each resultant simplex are recorded. Therefore, in the on-line phase, once the specific simplex which contains the required resultant wrench is found, the optimal contact forces can be calculated quickly by the combination of the primitive forces corresponding to the vertices of this simplex. The proposed two-phase runs very fast and the obtained contact forces are very close to the optimal solution.(5) Presents a method for unknown objects grasping using six-axis force/torque sensorsGeometrically unknown objects grasping is a challenging problem in robotic grasping. Mounting a six-axis force/torque sensor on the end of each robotic arm, we design a three-fingered robotic hand grasping system. Compared to previous work based on computer vision, our system is simpler and model reconstruction is not necessary. The position of the contact point and corresponding normal direction on the object can be calculated from the sensor information once contact occurs. Therefore, beside as an actuator to grasp the object, the robotic hand is a touching sensor and capable of "exploration" the shape of unknown objects through continual contact to find suitable force-closure grasp configurations.(6) Develops a novel three-fingered detours hand driven by shape memory alloyThe artificial muscle is the development trend of actuator for multi-fingered dexterous hand. Shape memory alloy is a special metal material with shape memory effect. It is a promising artificial muscle material because of its small size. It is also capable of integrating sensing, driving and transmission. In this thesis, a three-fingered dexterous hand named ZJUHand is developed. It is a compact hand with9-DOFs and the shape memory alloy is built into the fingers and palm. ZJUHand is the first SMA-driven dexterous hand in China, and filled related blank in domestic. Compared to the foreign SMA-driven dexterous hands, ZJUHand possess many advantages such as simple structure, compact size, light weight, low cost, etc. It is particularly suitable to install on the end of manipulator for fine operation with light load.