Dotan Di Castro
Rafael Advanced Defense Systems (Israel), Robert Bosch (India)
Papers
8
Total Citations
102
H-Index
5
About
Dotan Di Castro is a robotics and autonomous systems researcher whose work centers on robot manipulation, sensory perception, and multi-modal learning. He is best known for developing the InsertionNet framework, a scalable deep learning solution that enables robots to perform complex assembly insertion tasks across a wide variety of shapes and configurations — a longstanding industrial challenge. The original InsertionNet (2021, 49 citations) and its successor, InsertionNet 2.0 (2022, 21 citations), together represent a significant leap forward in practical robot assembly, introducing minimal-contact, multi-step insertion using multimodal and multi-view sensory input with minimal human intervention and no hand-crafted rewards. Beyond insertion, Di Castro has made notable contributions to robot grasping, developing hybrid motion-primitive strategies for cluttered environments and probabilistic object hierarchy representations through DUQIM-Net. His work on sim-to-real learning demonstrates an elegant approach to training robots simultaneously in simulation and the real world, improving sample efficiency and transfer performance. More recently, he has expanded into autonomous driving perception, proposing radar-camera fusion methods and few-shot instance segmentation for robotics. Across his portfolio, Di Castro consistently bridges the gap between theoretical machine learning and deployable real-world robotic systems, making his research highly relevant to both academic and industrial audiences.
Research Focus
Key Achievements
Top Papers
- 1InsertionNet - A Scalable Solution for Insertion49 citations · 2021
- 2
- 3
- 4
- 5Sim and Real: Better Together5 citations · 2021
- 6Robot Instance Segmentation with Few Annotations for Grasping4 citations · 2025
- 7RCF-TP: Radar-Camera Fusion With Temporal Priors for 3D Object Detection2 citations · 2024
- 8