Google Scholar
GitHub
LinkedIn
Near-field Interaction
Scene Interaction Parsing, Hand-Object Interaction Prediction, Manipulation Model.
Assistant Professor at Zhongguancun Academy
Xingyu Chen
I lead ZGCA HMI Lab, working on Interaction-centric Embodied AI.
I am currently an Assistant Professor at Zhongguancun Academy. Previously, I was an assistant research fellow at Peking University and IDEA-CVR, working closely with Prof. Lei Zhang.
Before joining PKU, I spent 4 years in industry at Kuaishou and Xiaobing, working closely with Dr. Baoyaun Wang. I received my Ph.D. from the Institute of Automation, Chinese Academy of Sciences in 2020 under Prof. Junzhi Yu, and my B.S. from Chengdu University of Technology in 2015.
For Ph.D. students and research interns
Get in touch
If you are interested in ZGCA and my research directions, please contact me.
Research Areas
My work sits at the intersection of computer vision, robotics, graphics, and human-machine interaction.
Spatial Interaction
Spatiotemporal State Representation, Controllable World Model, Navigation Model.
Social Interaction
Human-Robot Collaboration Framework, Human-Robot Social Interaction Framework, Multi-Embodied Agents.
Selected Publications
Citation counts: Google Scholar, automatically updated via Shields.io (cached).
PINT: A Unified Model with Shared Predictive Intent for Robot Manipulation
PINT is a unified model that leverages shared predictive intent for robot manipulation tasks.
4D Latent Cache for Efficient Conditioning in Dynamic World Modeling
4LC reuses cached visual representations across time and viewpoints to condition multi-task video generation efficiently.
HyrCAP: Hybrid Rank-Calibration of Action Proposals for Temporal Event Understanding
Hybrid propose rank-calibration method for of action proposal and semantic understanding.
Hand-Object Interaction in the Age of Large Foundation Models: Reconstruction, Generation, and Embodied Transfer
A systematic survey of foundation-model priors for HOI reconstruction, generation, and embodied transfer.
PriorVLA: Prior-Preserving Adaptation for Vision-Language-Action Models
Prior knowledge maintenance and efficient downstream fine-tuning in VLA models.
World-Ego Modeling for Embodied Video Generation in Long-Horizon Navigation-Manipulation Tasks
A video-based world-ego modeling framework for long-horizon evolution in hybrid embodied tasks.
VLingNav: Embodied Navigation with Adaptive Reasoning and Visual-Assisted Linguistic Memory
VLingNav combines adaptive reasoning with visual-assisted linguistic memory for persistent cross-modal semantic memory in long-horizon navigation.
Detect Anything via Next Point Prediction
A unified framework for point-based visual cognition based on LLMs.
Rex-Thinker: Grounded Object Refering via Chain-of-Thought Reasoning
Object referring is reformulated as a Chain-of-Thought reasoning task that verifies candidate object regions step by step.
HandOS: 3D Hand Reconstruction in One Stage
An end-to-end framework for hand detection, 2D pose estimation, and 3D mesh reconstruction with a unified hand representation.
DINO-X: A Unified Vision Model for Open-World Object Detection and Understanding
A unified framework for visual perception.
HAVE-FUN: Human Avatar Reconstruction from Few-Shot Unconstrained Images
A framework for creating human avatars from few-shot unconstrained images.
Portrait4d: Learning one-shot 4d head avatar synthesis using synthetic data
A framework for creating human avatars from few-shot unconstrained images.
Mimic3D: Thriving 3D-Aware GANs via 3D-to-2D Imitation
A 3D-aware GAN that can directly render high-resolution images.
Hand Avatar: Free-Pose Hand Animation and Rendering from Monocular Video
A neural hand rendering framework with self-occluded illumination.
MobRecon: Mobile-Friendly Hand Mesh Reconstruction from Monocular Image
A single-view hand mesh reconstruction framework with accuracy, fast inference, and temporal coherence.
Camera-Space Hand Mesh Recovery via Semantic Aggregation and Adaptive 2D-1D Registration
A camera-space mesh recovery method that separates root-relative mesh recovery and root recovery.
A soft manipulator for efficient delicate grasping in shallow water: Modeling,control,and real-world experiments
A marine robot with a soft manipulator for efficient delicate grasping in shallow sea.
Towards real-time advancement of underwater visual quality with GAN
GAN-RS elevates underwater visual quality for real-time robotic perception.
Complete Publication Record
Browse first- or corresponding-author papers, co-author papers, and book in the redesigned full publication page.