Nguyen T.T. Nguyen

Computer Graphics · 3D Vision · Digital Humans for Robotics

prof_pic.jpg

Postdoctoral Researcher

Advanced Intelligent Space Lab

Ritsumeikan University · Osaka, JP

I build photorealistic, animatable digital humans and put them where people can interact with them: on screens, on holographic displays, and on a patient-simulator robot used to train medical and elderly-care staff.

I’m a postdoctoral researcher at the Advanced Intelligent Space Lab, Ritsumeikan University, where I also completed my Ph.D. (2026) as a MEXT scholar. My current work is multi-view 3D Gaussian Splatting: a pipeline that goes from pose-free sparse-view capture to real-time rendering of 4D head avatars. Earlier, I worked on generative facial behavior, teaching avatars how to listen, react, and express pain.

Before research, I spent two years as a backend/AI engineer at VNG Corporation, building ad systems that served millions of requests per second. That background still shapes how I work: I validate methods early on real, noisy data, and I care about whether a model fits the hardware and latency budget it will actually run on.

8peer-reviewed papers
2international awards
IROS '25 · REACT '24
3display targets
screen · hologram · robot
5 yrsindustry R&D
collaborations in Japan

research at a glance

One pipeline, from capturing a real face to giving it a body.

  1. 01 · capture

    Pose-free sparse-view 4D capture

    Calibration-free multi-view reconstruction of animatable 3DGS head avatars. Coarse feed-forward poses are refined with pycolmap and supervise a deformable cross-view point-map network.

    UR '26

    3DGSpycolmapmulti-view geometryPyTorch

  2. 02 · track

    Temporally stable 3DMM tracking

    Monocular and multi-view stereo face trackers with learnable camera parameters. They turn in-the-wild video into 3DMM animation datasets with low latency.

    SII '24

    3DMMOpenCVC++

  3. 03 · behave

    Generative facial behavior

    Avatars that listen and react to speech and vision. Methods include sparse keyframe motion tokens, FSQ facial tokenizers, and diffusion models that express pain.

    TCSS '25 IROS '25 ★ FG '24 ★ ISR survey

    TransformersdiffusionVQ / FSQ

  4. 04 · embody

    Real-time rendering & embodiment

    Custom CUDA kernels speed up 3DGS rendering on a consumer GPU. Avatars stream to Looking Glass holographic displays and drive a robotic shoulder-arm for range-of-motion training in VR.

    ICEIC

    CUDAOpenGLLooking Glass SDKROS 2Unity

news

Oct 2026 Started as a Postdoctoral Researcher at Ritsumeikan University, continuing work on 3DGS digital humans for care training.
Sep 2026 Completed my Ph.D. in Information Science and Engineering at Ritsumeikan University.
Oct 2025 PainDiffusion received the Best Entertainment and Amusement Paper Award at IEEE/RSJ IROS 2025.
May 2024 Our team won the REACT 2024 Challenge at IEEE FG 2024 with an FSQ-based facial tokenizer for dyadic reaction generation.
Jan 2024 Presented our latency-aware 3DMM face-tracking work at IEEE/SICE SII 2024.

selected publications

  1. UR
    Pose-Free Sparse View 4D Face Capture with 3D Gaussian Splatting
    Tri Tung Nguyen Nguyen*, Y. Hosoda , and Joo-Ho Lee
    In 2026 23rd International Conference on Ubiquitous Robots (UR), 2026
  2. ISR
    A Survey on Generative Non-verbal Facial Behavior for Highly Realistic Embodied Agents
    Tri Tung Nguyen Nguyen*, Q Tien Dam, D Tuan Tran, and 1 more author
    Intelligent Service Robotics, 2026
  3. TCSS
    sfms.png
    When Less is More: A Sparse Facial Motion Structure for Listening Motion Learning
    Tri Tung Nguyen Nguyen*, Q Tien Dam, D Tuan Tran, and 1 more author
    IEEE Transactions on Computational Social Systems, 2025
  4. IROS
    PainDiffusion: Learning to Express Pain
    Quang Tien† Dam, Tri Tung Nguyen Nguyen*†, Yuki Endo, and 2 more authors
    In 2025 IEEE/RSJ International Conference on Intelligent Robots and Systems (IROS), 2025