Papers
6
Total Citations
109
H-Index
4
About
Dongchen Zhu is a leading researcher in robotic perception and 3D scene understanding, whose work bridges the gap between semantic mapping, object pose estimation, and visual odometry. His most influential contribution is the development of a novel RGB-D semantic segmentation framework combined with label-oriented voxelgrid fusion, enabling accurate 3D semantic mapping from RGB-D scans—a foundational technology for task-driven robots (67 citations). Zhu has also pioneered methods for dynamic obstacle rejection in 3D map updating, using bidirectional searching with KD trees to eliminate spurious trails from moving objects, significantly improving mapping robustness in dynamic environments. His recent work on category-level 6D object pose estimation introduces structural discrepancy awareness to handle shape variations in unseen objects, advancing robot grasping and augmented reality applications. Zhu’s research extends to self-supervised visual-inertial odometry, where he developed a scale recovery method that leverages inertial data more effectively, and cascade contour-enhanced panoptic segmentation for robotic vision. With a growing citation impact and a focus on bionic binocular robots and real-world deployment, Zhu’s contributions are shaping the next generation of intelligent robotic systems capable of robust, fine-grained scene perception.
Research Focus
Key Achievements
Top Papers
- 1
- 2Dynamic Obstacles Rejection for 3D Map Simultaneous Updating14 citations · 2018
- 3
- 4
- 5
- 6Self-supervised Scale Recovery for Decoupled Visual-inertial Odometry3 citations · 2023