- Present
Research Intern, JingDong Future Academy
Beijing, China · Embodied AI, world-action models, and world models.
Mentor: Dr. Jun Cen
Yifan Ye
I am a research intern at JingDong Future Academy, mentored by Dr. Jun Cen, working on embodied AI, world-action models, and world models. Previously, I was a research intern at BAAI, mentored by Prof. Shanghang Zhang, and a research assistant at Hamad Bin Khalifa University, supervised by Dr. Zhihe Lu.
I received my M.E. and B.E. in Electronic Information Engineering from Guangdong University of Technology in 2025 and 2022, respectively, under the supervision of Prof. Jing Chen.
My research interests include embodied AI, world-action models, vision-language-action models, tactile manipulation, and efficient robot learning.
Research Experience
-
Research Intern, Beijing Academy of Artificial Intelligence (BAAI)
Beijing, China · Vision-language-action and world-action models.
Mentor: Prof. Shanghang Zhang
-
Research Assistant, Hamad Bin Khalifa University
Doha, Qatar
Supervisor: Dr. Zhihe Lu
-
Visiting Student, Southern University of Science and Technology (SUSTech)
Shenzhen, China · Computer Science
Publications
Full list on Google Scholar
Data Pyramid for Embodied Manipulation
arXiv preprint arXiv:2607.24744, 2026
A data-centric survey that organizes embodied manipulation data into a five-level pyramid and connects heterogeneous data recipes to embodied foundation models and open research challenges.
Self-evolved Imitation Learning in Simulated World
IEEE Robotics and Automation Letters (RA-L), 2026
SEIL enables few-shot policies to improve through simulator rollouts, dual-level augmentation, and informative-trajectory selection.
Token Expand-Merge: Training-Free Token Compression for Vision-Language-Action Models
IEEE Robotics and Automation Letters (RA-L), 2026
TEAM-VLA accelerates VLA inference without retraining by expanding informative tokens and merging redundant ones under action-aware guidance.
Dream-Tac: A Unified Tactile World Action Model for Contact-Rich Robot Manipulation
ACM Multimedia, 2026
Dream-Tac jointly models actions, future vision, and tactile dynamics for contact-rich manipulation through contact-aware fusion and efficient inference.
Efficient-WAM: A 1B-Parameter World-Action Model with Low-Cost Future Imagination
Technical Report, 2026
Efficient-WAM combines a compact video expert, sparse future latents, and asymmetric denoising to reduce the cost of future imagination.
Learning Cross-Domain Representations by Vision Transformer for Unsupervised Domain Adaptation
Neural Computing and Applications, 2023
A self-training UDA method combining cross-domain patch aggregation with centroid-alignment loss for vision Transformers.
* Equal contribution. † Project leader. ‡ Corresponding author.
Selected Honors
- Starlight Excellence Award, DIGIX Global Campus AI Algorithm Elite Competition, Huawei.
- First-Class Scholarship, Guangdong University of Technology.
- Excellence Program Scholarship, Guangdong University of Technology.