#### Thesis Format

Integrated Article

#### Degree

Doctor of Philosophy

#### Program

Electrical and Computer Engineering

#### Abstract

The demographic shift has caused labor shortages across the world, and it seems inevitable to rely on robots more than ever to fill the widening gap in the workforce. The robotic replacement of human workers necessitates the ability of autonomous grasping as the most natural but rather a vital part of almost all activities. Among different types of grasping, fingertip grasping attracts much attention because of its superior performance for dexterous manipulation. This thesis contributes to autonomous fingertip grasping in four areas including hand-eye calibration, grasp quality evaluation, inverse kinematics (IK) solution of robotic arm-hand systems, and simultaneous achievement of grasp planning and IK solution.

To initiate autonomous grasping, object perception is the first needed step. Stereo cameras are well-embraced for obtaining an object's 3D model. However, the data acquired through a camera is expressed in the camera frame while robots only accept the commands encoded in the robot frame. This dilemma necessitates the calibration between the robot (hand) and the camera (eye) with the main goal is of estimating the camera's relative pose to the robot end-effector so that the camera-acquired measurements can be converted into the robot frame. We first study the hand-eye calibration problem and achieve accurate results through a point set matching formulation. With the object's 3D measurements expressed in the robot frame, the next step is finding an appropriate grasp configuration (contact points + contact normals) on the object's surface. To this end, we present an efficient grasp quality evaluation method to calculate a popular wrench-based quality metric which measures the minimum distance between the wrench space origin ($\vec{0}_{6\times 1}$) to the boundary of grasp wrench space (GWS). The proposed method mathematically expresses the exact boundary of GWS, which allows to evaluate the quality of the grasp with the speed that is desirable in most robotic applications. Having obtained a suitable grasp configuration, an accurate IK solution of the arm-hand system is required to perform the planned grasp. Conventionally, the IK of the robotic hand and arm are solved sequentially, which often affects the efficiency and accuracy of the IK solutions. To overcome this problem, we kinematically integrate the robotic arm and hand and propose a human-inspired Thumb-First strategy to narrow down the search space of the IK solution. Based on the Thumb-First strategy, we propose two IK solutions. Our first solution follows a hierarchical IK strategy, while our second solution formulates the arm-hand system as a hybrid parallel-serial system to achieve a higher success rate. Using these results, we propose an approach to integrate the process of grasp planning and IK solution by following a special-designed coarse-to-fine strategy to improve the overall efficiency of our approach.

#### Summary for Lay Audience

The demographic shift has caused labor shortages across the world, and it seems inevitable to rely on robots more than ever to fill the widening gap in the workforce. The robotic replacement of human workers necessitates the ability of autonomous grasping as the most natural but rather a vital part of almost all activities. This thesis contributes to the improvement of the accuracy and efficiency of the overall autonomous grasping process.

The target object's model is often needed to start the grasping process. Stereo cameras are well-embraced to reconstruct the object's 3D model. It is necessary to express the data acquired through a camera in the robot coordinate frame for commanding the robot with camera-acquired data. This requires estimating the camera's pose (position+orientation) relative to the robot's end-effector (the hand), which is well-known as the hand-eye calibration problem. We first study the hand-eye calibration problem for stereo cameras and achieve accurate calibration results for stereo cameras by formulating the hand-eye calibration problem as a point set matching problem. With the object's 3D model expressed in the robot coordinate system, we then study the efficient evaluation of the general capability of one grasp configuration (contact positions + contact directions), which can be accumulated to significantly expedite the evaluation process when numerous grasp configurations are involved. Once a grasp configuration is satisfactory, the inverse kinematics (IK) of the robotic arm-hand system needs to be solved to achieve this grasp in practice. Conventionally, the robotic hand's and arm's IK are solved sequentially, which may be inefficient and inaccurate. To release this potential limitation, we regard the robotic arm and hand as an integrated system and study the IK problem of integrated arm-hand systems. We notice that the sequential procedure of finding the desired grasp configuration (i.e., grasp planning) and solving the IK affects the overall efficiency. To improve the overall efficiency of our approach, we reorganized and intertwined the process of grasp planning and IK solution to solve these two problems simultaneously.

COinS