Stephen Tyree
Papers
10
Total Citations
172
H-Index
6
About
Stephen Tyree is a computer vision and robotics researcher whose work sits at the intersection of perception, pose estimation, and robot manipulation. He has made significant contributions to category-level 6-DoF object pose estimation, most notably through his single-stage keypoint-based method that enables robots to recognize and localize objects without requiring textured CAD models for each instance — a critical advance for real-world deployment. His widely cited work (49 citations) represents a meaningful shift from instance-level to category-level understanding in robotic vision. Tyree's research spans the full robotic perception pipeline: from sim-to-real transfer for precise grasping (29 citations), to multi-view scene understanding for manipulation (25 citations), to collision-free control using only RGB inputs. His HANDAL dataset (30 citations) provides a valuable benchmark specifically tailored to manipulable objects with pose annotations and affordance labels, directly addressing gaps in robotics-ready datasets. More recently, his RoboSpatial project (15 citations) targets spatial reasoning in vision-language models, reflecting a forward-looking interest in foundation models for robotics. Across his body of work, Tyree consistently bridges the gap between theoretical computer vision and practical robotic systems operating in unstructured environments.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4Multi-view Fusion for Multi-level Robotic Scene Understanding25 citations · 2021
- 5
- 6
- 7
- 8How to Close Sim-Real Gap? Transfer with Segmentation!6 citations · 2020
- 9
- 10