CS Events
Qualifying ExamSafe Object Rearrangement in Confined Spaces Under Visibility Constraints |
|
||
Friday, February 10, 2023, 02:00pm - 04:00pm |
|||
Abstract:
Robotic object rearrangement in confined spaces, such as shelves, is active research due to multiple applications in logistics and service robotics. Nevertheless, the majority of efforts do not explicitly consider visibility constraints, such as occlusions, or only work for simplified object models, such as cylinders. Some recent efforts in the area of “mechanical search” use machine learning and provide a practical framework for object manipulation under occlusions, but they are typically not able to provide theoretical guarantees.
This talk formulates general, occlusion-aware multi-object manipulation tasks and focuses on defining a theoretical framework that allows for performance guarantees. The framework assumes the use of RGBD data and object segmentation masks but does not assume access to 3D object models. To guarantee the safe rearrangement of objects in the general case, the effort focuses on guaranteeing the safe reconstruction of all objects in confined spaces under occlusions. A concrete implementation of the general framework for monotone instances is shown to outperform random and greedy baselines on randomly generated scenes in simulation. Even for cluttered scenes with realistic objects, the proposed algorithm significantly outperforms the baselines and maintains a high success rate across experimental conditions.
The talk will conclude with future directions that can improve the efficiency of such solutions while maintaining safety and probabilistic completeness guarantees.
Speaker: Yinglong Miao
Location : 1 Spring Street - Room 319
Committee:
Professor Kostas Bekris (Advisor)
Professor Jingjin Yu
Professor Jie Gao
Professor He Zhu
Event Type: Qualifying Exam
Abstract: See above
Organization:
Rutgers University
School of Arts & Sciences
Department of Computer Science
Contact Professor Kostas Bekris