A5061794221- Robot Manipulation and Learning
- Human Pose and Action Recognition
- Hand Gesture Recognition Systems
- Advanced Neural Network Applications
- Soft Robotics and Applications
- Image and Object Detection Techniques
- Advanced Vision and Imaging
- Robotics and Sensor-Based Localization
Dalian University of Technology
2020-2024
3D object detection is an important scene understanding task in autonomous driving and virtual reality. Approaches based on LiDAR technology have high performance, but expensive. Considering more general scenes, where there no data the datasets, we propose a approach from stereo vision which does not rely either as input or supervision training, solely takes RGB images with corresponding annotated bounding boxes training data. As depth estimation of key factor affecting performance...
Category-level 6D pose estimation can be better generalized to unseen objects in a category compared with instance-level estimation. However, existing category-level methods usually require supervised training sufficient number of annotations which makes them difficult applied real scenarios. To address this problem, we propose self-supervised framework for paper. We leverage DeepSDF as 3D object representation and design several novel loss functions based on help the model predict poses...
To assist or replace human beings in completing various tasks, research on the functional grasp synthesis of dexterous hands with high degree-of-freedom (DoF) is necessary and challenging. The requires not only that stable but more importantly facilitates manipulation after grasping. Such work still relies manual annotation when collecting data. this end, we propose a category-level multi-fingered transfer framework, which need to label hand-object contact relationship parts one object, then...
Successful grasp is an important and long-standing issue for robots to interact with the real world. Most recent studies have devoted more attention stable rather than functional grasp, which cannot guarantee task-oriented postgrasp manipulation. To achieve human-like a semantic representation of hand-object interaction introduced without labeling 3D hand poses, novel coarse-to-fine generation network designed model this interaction. First, coarse generated by combining global pose type....