Robotic perception and grasp in unstructured environments

• Main Author: Dong, Huixu
• Other Authors: Chen I-Ming
• Format: Theses and Dissertations
• Language: English
• Published: 2019
• Subjects: DRNTU::Engineering::Mechanical engineering::Robots
• Online Access: https://hdl.handle.net/10356/89885; http://hdl.handle.net/10220/47731
• Institution: Nanyang Technological University

Perception and grasp are crucial capabilities for robots to perform desired services in unstructured environments. This thesis presents important insights and concepts related robotic perception and grasp in unstructured scenarios. In this thesis, three specific problems related to this topic have been studied, viz., tracking of cylindrical objects, grasping of static and dynamic cylindrical objects based on the proposed ellipse detection, pose estimation of multiple objects and occluded objects in cluttered environments and the optimal design of under-actuated robotic gripper for realizing stable grasps. For the problem of the ellipse detection, two critically important problems have been addressed. The proposed detector provides either faster or more accurate ellipse detection results than the current state-of-the-art methods, irrespective of challenging scenarios such as occluded or overlapping ellipses. For the problem of pose estimation of objects, we propose a highly efficient learning approach integrated by the contextual information to estimate pose of the textured or texture-less objects for grasping purposes in a cluttered environment where the objects might be partially occluded. It has been indicated that the proposed method is superior to several state-of-the-art works. The proposed perception algorithms impose the constraints in the scenarios where the severe occlusions result in the lack of visibility. For the problem of the optimal design of under-actuated robotic gripper, the mathematical model between the active and contact forces has been expressed and the geometric model of transmission characteristics determined by the tendon routes for reducing the resistance has been explored for determining the dimension parameters of the gripper. Practical experiments are performed by the designed gripper to validate the proposed designed approach. The utility of these algorithms has been shown using several series of robotic grasp experiments with successful rates of over 80% in various difficult scenarios, including tracking cylindrical objects, grasping static and dynamic cylindrical objects, grasping textured and texture-less by estimating poses of multiple objects and occluded objects.