Xinchao Song
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Research Subjects
Xinchao Song has a total of 4 co-authors. Here are the top co-authors by number of shared publications:
- Natasha Kholgade Banerjee
Wright State University
2 shared publications - Sean Banerjee
Wright State University
2 shared publications - James Cox
Department of Computer Science and Engineering, Wright State University, Dayton, OH, USA
1 shared publication - Mingjun Li
University of Hartford, West Hartford, Connecticut, USA
1 shared publication
Latest Publications
- Open Access
- 24 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Data Descriptions
- Vol. 3, 203-211
Descriptor: Real-World 3-D Object, Human Giver, Human Receiver, and Sampled Robot Grasps Dataset (HOH-Grasps)
With the spread of robotic systems that perform close-proximity interactions with humans, it is important to have datasets that enable creation of robot object grasp and manipulation algorithms that show awareness of how humans interact with objects. In this paper, we describe HOH-Grasps, a large-scale dataset containing data on robot grasps synthetically sampled from real-world 3D object point clouds combined with point clouds of human giver and receiver hands acquired from real-world captures of human-human object handover interactions. HOH-Grasps contains object, giver hand, and receiver hand point clouds from 2,720 scenes, spanning 40 giver-receiver participant pairs and 136 objects. The robotic grasp data consists of 2,047,492,800 antipodal, parallel-jaw grasps over all 136 objects, of which 73,682,572 are valid, i.e., have no object collisions, and 25,432,094 total grasps are high-quality, i.e., have a friction coefficient ≤ 0.5 to achieve force closure. When object grasps are transformed to all 2,720 scenes, the dataset provides 2,316,434,197 valid grasps, i.e., grasps without collision with the object or tabletop plane, and 802,228,990 high-quality grasps. The dataset provides value in enabling training and evaluation of human- and object-aware robot grasp and manipulation algorithms, as well as broad-range tasks of interest in robotics and vision, such as object classification and pose estimation, to enable robots to engage in physical human-robot interaction tasks such as handover and collaborative assembly. The dataset is shared on HuggingFace at: https://huggingface.co/datasets/tars-home/HOH-Grasps.
- 24 March 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Data Descriptions
- Vol. 3, 203-211
Open AccessDescriptor: Real-World 3-D Object, Human Giver, Human Receiver, and Sampled Robot Grasps Dataset (HOH-Grasps)
With the spread of robotic systems that perform close-proximity interactions with humans, it is important to have datasets that enable creation of robot object grasp and manipulation algorithms that show awareness of how humans interact with objects. In this paper, we describe HOH-Grasps, a large-scale dataset containing data on robot grasps synthetically sampled from real-world 3D object point clouds combined with point clouds of human giver and receiver hands acquired from real-world captures of human-human object handover interactions. HOH-Grasps contains object, giver hand, and receiver hand point clouds from 2,720 scenes, spanning 40 giver-receiver participant pairs and 136 objects. The robotic grasp data consists of 2,047,492,800 antipodal, parallel-jaw grasps over all 136 objects, of which 73,682,572 are valid, i.e., have no object collisions, and 25,432,094 total grasps are high-quality, i.e., have a friction coefficient ≤ 0.5 to achieve force closure. When object grasps are transformed to all 2,720 scenes, the dataset provides 2,316,434,197 valid grasps, i.e., grasps without collision with the object or tabletop plane, and 802,228,990 high-quality grasps. The dataset provides value in enabling training and evaluation of human- and object-aware robot grasp and manipulation algorithms, as well as broad-range tasks of interest in robotics and vision, such as object classification and pose estimation, to enable robots to engage in physical human-robot interaction tasks such as handover and collaborative assembly. The dataset is shared on HuggingFace at: https://huggingface.co/datasets/tars-home/HOH-Grasps.
- Open Access
- 9 January 2026
- Published by Elsevier in Data in Brief
- Vol. 65, 112450
Dataset of RGB-D images of object collections from multiple viewpoints with aligned high-resolution 3D models of objects
- 9 January 2026
- Published by Elsevier in Data in Brief
- Vol. 65, 112450
Open AccessDataset of RGB-D images of object collections from multiple viewpoints with aligned high-resolution 3D models of objects