Danfei Xu

About

Danfei Xu is a robotics and machine learning researcher whose work spans robot perception, manipulation, task planning, and imitation learning. He is perhaps best known for **DenseFusion** (2019, 1,121 citations), a highly influential framework for 6D object pose estimation that pioneered iterative dense fusion of RGB and depth data, significantly advancing robotic scene understanding in cluttered environments. His early contributions include tactile-based object identification using Bayesian exploration (2013) and trajectory optimization for deformable object manipulation (2015), demonstrating a career-long commitment to dexterous robot interaction with the physical world. Xu has also made substantial contributions to robot learning and generalization. His Neural Task Programming framework (2018) introduced hierarchical few-shot learning from demonstrations, while GTI and follow-on imitation learning work pushed boundaries in long-horizon task generalization. More recently, his ProgPrompt system (2023, 508 citations) helped establish the paradigm of using large language models for situated robot task planning, earning rapid recognition from the research community. His involvement in the landmark Open X-Embodiment collaboration (2024) further reflects his commitment to large-scale, generalizable robot learning. Collectively, Xu's research shapes how robots perceive, reason about, and act in complex real-world environments.

Research Focus

Key Achievements

Top Papers

1. 1
   DenseFusion: 6D Object Pose Estimation by Iterative Dense Fusion

   1,121 citations · 2019
2. 2
   ProgPrompt: Generating Situated Robot Task Plans using Large Language Models

   508 citations · 2023
3. 3
   6-PACK: Category-level 6D Pose Tracker with Anchor-Based Keypoints

   159 citations · 2020
4. 4
   Tactile identification of objects using Bayesian exploration

   155 citations · 2013
5. 5
   Neural Task Programming: Learning to Generalize Across Hierarchical Tasks

   153 citations · 2018
6. 6
   Open X-Embodiment: Robotic Learning Datasets and RT-X Models : Open X-Embodiment Collaboration⁰

   119 citations · 2024
7. 7
   Folding deformable objects using predictive simulation and trajectory optimization

   107 citations · 2015
8. 8
   DenseFusion: 6D Object Pose Estimation by Iterative Dense Fusion

   93 citations · 2019
9. 9
   GTI: Learning to Generalize across Long-Horizon Tasks from Human Demonstrations

   86 citations · 2020
10. 10
    What Matters in Learning from Offline Human Demonstrations for Robot Manipulation

    70 citations · 2021

Key Collaborators

Contact & Links