Object manipulation and environment interaction are of great significance for intelligent robots, especially service robots working under unstructured household and office scenarios. This paper proposes a novel approach for categorical unseen object grasping and manipulation. Different from recently popular end-to-end reinforcement learning methods, we develop models for geometric primitive abstraction of target objects, and accordingly estimate their pose as well as generate task-orientated grasp points. Such design emphasizes visual perception in guiding robotic manipulation, thereby enhancing model interpretability and reliability during implementation. In addition, we also conduct object grasping experiments both under simulation and real-world settings, which further verify the effectiveness and superiority of our method.
Vision-Based Categorical Object Pose Estimation and Manipulation
Lect.Notes Computer
International Conference on Intelligent Robotics and Applications ; 2023 ; Hangzhou, China July 05, 2023 - July 07, 2023
2023-10-21
12 pages
Article/Chapter (Book)
Electronic Resource
English
Robotic Manipulation , Object Representation , Pose Estimation Computer Science , Artificial Intelligence , Software Engineering/Programming and Operating Systems , Computer Applications , Computer Communication Networks , Special Purpose and Application-Based Systems , User Interfaces and Human Computer Interaction
Vision-Based Categorical Object Pose Estimation and Manipulation
TIBKAT | 2023
|Vision Based Object Pose Estimation for Mobile Robots
British Library Conference Proceedings | 1994
|Vision based object pose estimation for mobile robots
NTRS | 1994
|