Papers
6
Total Citations
83
H-Index
5
About
Atabak Dehban is a roboticist whose research lies at the intersection of computer vision, machine learning, and cognitive robotics, with a focus on enabling robots to perceive, predict, and interact with their environments. His key contributions center on learning object affordances—the action possibilities offered by objects—using denoising auto-encoders, a foundational approach that has garnered 30 citations and supports capabilities like prediction and planning. Dehban has also advanced robotic grasping through his work on 3DSGrasp, which uses 3D shape completion to generate robust grasps from incomplete point cloud data, a practical solution for real-world manipulation. His development of a generic visual perception domain randomization framework for Gazebo (17 citations) addresses the challenge of sim-to-real transfer in deep learning, while his research on learning deep features from physical interactions (6 citations) and action-conditioned graph neural networks for soft robotic hand dynamics (5 citations) further demonstrates his versatility. Dehban’s work has been published in top venues like ICRA and IROS, and his comparative study of video prediction models for robotics underscores his commitment to benchmarking and reproducibility. With over 80 total citations, his research is shaping how robots learn from and act upon the physical world.
Research Focus
Key Achievements
Top Papers
- 1
- 23DSGrasp: 3D Shape-Completion for Robotic Grasp23 citations · 2023
- 3A generic visual perception domain randomisation framework for Gazebo17 citations · 2018
- 4Learning Deep Features for Robotic Inference From Physical Interactions6 citations · 2022
- 5
- 6