Title:  Multi-level Scene Understanding for Advanced Manipulation

Committee: 

Dr. Vela, Advisor

Dr. Yezzi, Chair

Dr. AlRegib

Abstract: The objective of the proposed research is to explore multi-level visual information for advanced robotics manipulation, which would enhance the robot's capability to operate in less-structured scenes. Recent researches mainly focus on the task of how to grasp. Deep learning has emerged as a promising approach for addressing these issues by detecting grasp representation associated with a parallel plate gripper. However, they still suffer from the insufficiency of training data and the limited grasp configurations. Beyond how to grasp, how to manipulate is a more demanding task. It requires the robot to be aware of target-centric information. This ability includes locating objects and their poses, also known as the 6-DoF pose estimation problem (i.e., 6 degrees of freedom, from 3D position + orientation). Accurate, real-time pose information of nearby objects in the scene would allow robots to engage in semantic interaction. Our preliminary researches focus on improving robotic manipulation with primitive shape recognition and category-level pose estimation. Our proposed work further extends the generality of object pose estimation to unseen objects and brings a broad impact on the applicability.