CS PhD Student · UCLA

Wenbo (Gordon) Hu

Hi! I am Wenbo. I'm a CS PhD student at University of California, Los Angeles advised by Prof. Kai-Wei Chang and Prof. Nanyun Peng. I also obtained my M.S. in CS from UCLA in 2024. Previously, I obtained my B.S. in Data Science at the Halicioglu Data Science Institute (HDSI) from University of California, San Diego in 2023. I'm fortunate to have worked with Prof. Zhuowen Tu and Prof. Hao Su during my undergraduate study.

I work on LLM, multimodal, and agents. Particularly, my past projects include 2D and 3D LLM/VLMs in multimodal understanding, reasoning, reconstruction, generation, agentic and embodied tasks. My long-term research goal is to build intelligent systems that can perceive, understand and interact with the complex digital and physical world.

CV ↗ GitHub Google Scholar LinkedIn Email Twitter (X) DBLP
Portrait of Wenbo Hu
01

Research

My primary research works include:

LLM/VLM, Agent

Pretraining / Continued Pretraining (Mid-training)

  • G2VLM: Geometry Grounded VLM
    CVPR 2026
  • BLIVA: General Multimodal LLM
    AAAI 2024

Post-Training / RL

  • Tev: Long-Horizon Agentic Environments with Progress Rewards
  • ARES: Multimodal Adaptive Reasoning
    ICLR 2026
  • OpenVLThinkerV2 / G2RPO: Gaussian GRPO
    COLM 2026

Memory & Efficiency

  • 3DLLM-Mem: Long-Term Memory for Embodied Agent
    NeurIPS 2025
  • MQT-LLaVA: Matryoshka Query Transformer
    NeurIPS 2024

Benchmarks

MMSI-Video-Bench MRAG-Bench VALOR-EVAL TemMed-Bench

I also enjoy working with training and inference infras. My works have been trained on various platforms, including GPU, Google TPU (e.g. 3DLLM-Mem) and AWS Trainium (e.g. OpenVLThinkerV2).

02

News

  • We release Tev-27B and Tev-320B, with Tev-320B being the best open-source model on TB 4.0-CM!
  • We release OpenVLThinkerV2 and introduce G2RPO!
  • A unified 3D reconstruction and spatial reasoning model G2VLM is accepted by CVPR 2026!
  • Glad to organize the 2nd Workshop on 3D-LLM/VLA at CVPR 2026 together with Profs. Leonidas Guibas, Fei-Fei Li, and Yejin Choi and other awesome organizing committee members!
  • 3DLLM-Mem is selected for the 🏆 Best Paper Award at CVPR 2025 Foundation Models Meet Embodied Agents Workshop!
  • MQT-LLaVA is accepted at NeurIPS 2024!
  • BLIVA has reached over 200 citations. Thank the community's interests!
03

Selected Publications

Full: here

04

Experience

  • Adobe Research

    Research Scientist Intern

    Mentor: Yong Jae Lee

    June 2026 - Present

  • Shanghai AI Lab

    Research Intern at Embodied AI Center

    Mentors: Jiangmiao Pang, Tai Wang

    May 2025 - Sept 2025

  • Google Research

    Research Collaboration

    Mentors: Idan Szpektor, Yonatan Bitton

    Nov 2024 - May 2025
05

Service

  • CVPR 2026 Workshop Secondary Organizer at the 2nd Workshop on 3D-LLM/VLA.
  • Reviewer for ICLR, NeurIPS, ICML, CVPR, ICCV, ECCV, ACL, EMNLP, NAACL, ACL Rolling Review
06

Awards

  • Amazon Trainium Fellowship
  • Best Paper Award at CVPR 2025 Foundation Models Meet Embodied Agents Workshop
  • UCLA CS Departmental Fellowship Award