Yangxiao Lu
Papers
6
Total Citations
38
H-Index
4
About
Yangxiao Lu is a robotics and computer vision researcher whose work centers on enabling robots to perceive and manipulate objects in unstructured, real-world environments. His primary contributions lie in **unseen object instance segmentation (UOIS)** — the challenging problem of identifying and delineating objects a robot has never encountered during training. Lu's most influential work introduces the Mean Shift Mask Transformer, which elegantly modernizes classical mean shift clustering within a transformer-based framework to dramatically improve segmentation of novel objects, accumulating 14 citations since its 2024 publication. He has also pioneered self-supervised learning approaches that leverage long-term robot interaction to iteratively refine segmentation models without manual annotation, reflecting a practical, closed-loop vision of robotic learning. His RISeg framework further advances the field by exploiting body frame-invariant features through active robot interaction. Beyond perception, Lu has contributed to reproducible robotics benchmarking through SceneReplica, providing the community with a standardized evaluation platform for pick-and-place manipulation. His more recent work on novel instance detection extends these capabilities to few-shot recognition scenarios. With over 35 cumulative citations across six publications, Lu is establishing himself as a thoughtful contributor bridging robot perception and autonomous manipulation research.
Research Focus
Key Achievements
Top Papers
- 1Mean Shift Mask Transformer for Unseen Object Instance Segmentation14 citations · 2024
- 2
- 3
- 4
- 5Mean Shift Mask Transformer for Unseen Object Instance Segmentation3 citations · 2022
- 6