Junhua Huang
Junhua (Harry) Huang at night under the northern lights
Junhua smiling on a sunny street, a Canon camera on a strap around his neck
Junhua laughing on a riverbank in autumn, holding up a large fish
Junhua in a yellow jacket waving, snowboard under his arm, on a lit trail between bare trees at night
Selfie of Junhua smiling on a steep clay bluff above the water at sunset

Junhua (Harry) Huang

I'm a master's student in Electrical and Computer Engineering at UCLA. I do research at VAIL with Prof. Bolei Zhou.

I work on robot learning with real hardware. So far that has meant a power wheelchair, a quadruped with a robot arm on its back, and Meta Aria glasses.

I'm looking for a full-time role or a PhD position in robot learning and robotics, starting in 2027.

Résumé (PDF) Email LinkedIn GitHub

  1. ChairNav

    VAIL, UCLA First author Jan 2026 – Present

    We taught a power wheelchair to drive long outdoor routes using a delivery robot's driving data and its rider's corrections. I built the cross-embodiment data pipeline and worked on the chair's hardware and onboard software.

    Project Paper Drive the model

  2. Cue the Flow

    VAIL, UCLA Co-author 2026

    We steer flow-matching policies for delivery tasks in the open world. I integrated the robot arm on a quadruped and deployed the models on it, with a Jetson handling the camera images and sending them to remote inference.

    CoRL 2026 Project Paper

  3. Picking a wheelchair goal with your eyes

    VAIL, UCLA Student Researcher 2026

    I'm working on letting a rider pick the wheelchair's goal with Meta Aria Gen 2 glasses: look at a spot on the floor, then pinch. So far I've only tested it offline, with one wearer.

  4. Part of Figure 1 of the ToolSmith paper: an agent's expanded tool library, including an auto-generated "Event In Image" tool, whose trajectory identifies a tour bus as linked to the 2010 Manila hostage crisis.

    ToolSmith

    UCLA NLP Group Prof. Kai-Wei Chang Dec 2025 – Mar 2026

    I helped build a framework that writes and checks reusable Python tools for multimodal search agents, based on their failed runs.

    Findings of EMNLP 2026 Project

  5. Part of Figure 1 of the S3 paper: five seizure video frames of an AI-generated patient, faces blurred, labelled blank stare, head turning, vocalization, tonic and clonic, above a colour-coded timeline and a short semiology report.

    Seizure-Semiology-Suite (S3)

    Seizure Lab, ECE, UCLA Student Researcher Aug – Oct 2025

    I built a clinical seizure-semiology dataset and served three vision-language models with vLLM for the S3 benchmark.

    ICML 2026 Spotlight Paper

  6. Part of Figure 1 of the acoustic-highlighting paper: input audio passes through magnitude and waveform encoders, a latent highlighting transformer and decoders, conditioned on a video frame through CLIP and on a text feature module.

    Acoustic highlighting & co-speech gestures

    University of Rochester Prof. Chenliang Xu 2024 – 2025

    I'm first author of a paper on vision-language-guided acoustic highlighting. Before that, I worked on co-speech gesture generation for GestureLSM.

    Acoustic highlighting GestureLSM

  7. Figure 1 of the Video-LMM post-training survey: a flow from a pretrained LLM and encoders through supervised fine-tuning, reinforcement learning and test-time scaling, with small thumbnails of related datasets and methods.

    Skeleton-to-image & Video-LMM survey

    University of Rochester Prof. Jiebo Luo 2024 – 2025

    I generated animal images from skeletons with Stable Diffusion and preference optimization. I also contributed to a survey on Video-LMM post-training.

    Video-LMM survey

  8. Selfie of Junhua in the Farsee2 office, smiling in the foreground, with colleagues at their desks and monitors behind him.

    City-scale Gaussian Splatting

    Farsee2 Technology ML Intern Jul – Sep 2024

    I cut down floating artifacts in city-scale reconstructions with incidence-angle weighting and depth/normal supervision, and split the preprocessing across machines.

Papers

All publications

  1. Under submission

    ChairNav: Cross-Embodiment Pre-training and Personalization for Long-Horizon Wheelchair Navigation

    J. Huang*, Z. Liu*, H. He, X. Zhang, B. Zhou

    PaperProject

  2. CoRL 2026

    Cue the Flow: Steering Flow-Matching Policies for Open-World Delivery Manipulation

    H. Wang, G. Galimi, J. Huang, S. Song, W. Wu, Y. Yan, B. Zhou

    PaperProject

  3. Findings of EMNLP 2026

    ToolSmith: An Automatic Tool-Generation Framework for Multimodal Search Agents

    R. Sun, J. Huang, L. Zancato, M. M. Islam, K.-W. Chang

    Project

  4. ICML 2026 Spotlight

    Seizure-Semiology-Suite (S3): A Clinically Multimodal Dataset, Benchmark, and Models for Seizure Semiology Understanding

    L. Zhang, T. Monsoor, …, J. Huang, …, R. Mazumder, V. Roychowdhury

    Paper

  5. ICCV 2025

    GestureLSM: Latent Shortcut based Co-Speech Gesture Generation with Spatial-Temporal Modeling

    P. Liu, L. Song, J. Huang, H. Liu, J. Zhu, C. Xu

    Paper

  6. arXiv 2026

    Semantic Visually-Guided Acoustic Highlighting with Large Vision-Language Models

    J. Huang, C. Huang, C. Xu

    Paper

  7. arXiv 2026

    GestureFAR: Streaming Co-Speech Gesture Generation with Flow Autoregression

    P. Liu, H. Liu, J. Luo, J. Huang, C. Zou, L. Song

    Paper

* Equal contribution.

Education

Full CV
  1. University of California, Los Angeles M.S. Electrical and Computer Engineering Expected 2027
  2. University of Rochester B.S. Computer Science & B.S. Business · GPA 3.82 May 2025