Xiang Yu

About

Xiang Yu is a prominent computer vision and robotics researcher whose work sits at the intersection of object perception, scene understanding, and autonomous systems. His research spans three interconnected areas: 6D object pose estimation, multi-object tracking, and robot perception in unstructured environments. Yu's most influential contributions include the DeepIM framework for iterative 6D pose refinement (784 combined citations) and PoseCNN, a convolutional architecture for pose estimation in cluttered scenes, which together have reshaped how robots perceive and interact with objects. His 2015 work on online multi-object tracking via decision-making (716 citations) remains a landmark contribution to real-time video analysis, with direct applications in autonomous driving and navigation. Beyond pose estimation, Yu has pioneered self-supervised learning approaches that reduce costly manual annotation, enabling robots to learn from their own interactions. His work on unseen object instance segmentation addresses the critical challenge of generalizing to novel environments, while PoseRBPF advances continuous 6D pose tracking using probabilistic filtering. With over 2,300 citations across his top works, Yu's research consistently bridges theoretical innovation and practical robotics deployment, making him a significant voice in embodied AI and robot autonomy.

Research Focus

Key Achievements

Top Papers

1. 1
   Learning to Track: Online Multi-object Tracking by Decision Making

   716 citations · 2015
2. 2
   DeepIM: Deep Iterative Matching for 6D Pose Estimation

   581 citations · 2018
3. 3
   DeepIM: Deep Iterative Matching for 6D Pose Estimation

   203 citations · 2019
4. 4
   Self-supervised 6D Object Pose Estimation for Robot Manipulation

   198 citations · 2020
5. 5
   PoseRBPF: A Rao–Blackwellized Particle Filter for 6-D Object Pose Tracking

   166 citations · 2021
6. 6
   PoseCNN: A Convolutional Neural Network for 6D Object Pose Estimation in Cluttered Scenes

   131 citations · 2017
7. 7
   DA-RNN: Semantic Mapping with Data Associated Recurrent Neural Networks

   127 citations · 2017
8. 8
   Unseen Object Instance Segmentation for Robotic Environments

   119 citations · 2021
9. 9
   Scaling Local Control to Large-Scale Topological Navigation

   61 citations · 2020
10. 10
    Learning RGB-D Feature Embeddings for Unseen Object Instance Segmentation

    45 citations · 2020

Key Collaborators

Contact & Links