Brian Shu
Brian Shu

Brian Shu

M.S. Computer Science, Purdue · CoRAL Lab

I’m Brian Shu, an M.S. student in Computer Science at Purdue University, working in the Cognitive Robot Autonomy & Learning (CoRAL) Lab with Prof. Ahmed H. Qureshi. I work on robot learning for long-horizon manipulation: imitation learning, post-training vision-language-action (VLA) models, and the benchmarks that tell you whether a policy actually works.

Right now that is a dishwasher-loading policy in NVIDIA Isaac Lab. Before Purdue I built the subtask environments and asset pipeline for Fiatlux, a humanoid ladder-climbing benchmark under review at ICRA 2027, and a force-monitored expert for cable insertion on a UR5.

I’m looking for a robotics research internship for Summer 2027 and can work full-time from May 10 to August 20, 2027. If your lab or team works on manipulation, imitation learning or evaluation, I would like to hear from you.

Projects

Four projects, each with its own page: what I built, media from the simulator, and the code.

  1. A simulated Unitree G1 humanoid standing on an A-frame ladder to reach a ceiling light fixture, and holding a light bulb beside the ladder.

    2026 · Isaac Lab · Unitree G1 · Benchmark · USD assets

    Fiatlux

    A long-horizon benchmark in which a simulated Unitree G1 humanoid climbs a ladder and replaces a light bulb, built in NVIDIA Isaac Lab.

  2. Isometric render of the simulated dishwasher with its door open and both racks pulled out, loaded with cups, a red mug and red bowls, with a cutlery basket in the lower rack.

    2026 · Isaac Lab · Manipulation · Planning · Imitation learning

    Dishwasher loading

    A general dishwasher-loading manipulation policy in NVIDIA Isaac Lab, built on simulated dish assets I released publicly.

  3. 2026 · ROS 2 · Gazebo · LeRobot · ACT · Diffusion Policy

    Cable insertion

    A force-monitored expert policy and an imitation-learning pipeline for cable insertion on a UR5, built for the AI for Industry Challenge.

  4. 2025 · VLM · Synthetic data · Qwen2.5-VL · Fine-tuning

    DCVLR

    Data Curation for Visual-Language Reasoning, a pipeline that generates synthetic multimodal question-answer pairs with reasoning steps and uses them to fine-tune Qwen2.5-VL.

Publications

  1. The G1 crouching beside an A-frame ladder and gripping it to move it under the ceiling fixture.

    arXiv 2026 · under review at IEEE ICRA 2027

    Fiatlux: A Long-Horizon Benchmark for Humanoid Ladder Climbing and Light-Bulb Replacement

    P. Bushuyeu*, Y. Chen*, A. Nikolaev*, B. Shu*, I. Molybog · *equal contribution, alphabetical

    A 12-subtask benchmark in NVIDIA Isaac Lab in which a simulated Unitree G1 humanoid moves a ladder, climbs it and replaces a ceiling light bulb.

Experience

  • CoRAL Lab, Purdue — Graduate ResearcherWest Lafayette · Aug 2026 – present

    Building a general dishwasher-loading manipulation policy in NVIDIA Isaac Lab; released the simulated dish assets.

  • HawAII, UH Mānoa — Research AssistantHonolulu · Mar – Sep 2026

    Built the 12 subtask environments and asset pipeline for the Fiatlux benchmark, and the force-monitored expert and imitation-learning pipeline for the AI for Industry Challenge.

  • Hyundai Card — Data Science InternSeoul · Jun – Jul 2025

    RAG pipeline for text-to-SQL with self-consistency decoding; migrated to an open-source model while keeping 80% execution accuracy.

  • UH Mānoa — Software Verification DeveloperHonolulu · May 2024 – May 2025

    Isabelle/HOL proofs of functional correctness, following the seL4 and RT-11 verification efforts.

  • UH Mānoa — Research AssistantHonolulu · Jun – Aug 2023

    Efficient simulation of a biological reaction model, with 2D and 3D visualization of the results.

More work

  • Jane Street Real-Time Market Data ForecastingKaggle

    Supervised autoencoder + MLP; ranked 440 of 3,000 (top 15%).

  • Spinner Dolphin Photo Grading ProtocolComputer vision

    Multi-stage model that detects, rates and classifies spinner dolphins by fin distinctiveness.

  • BirdCLEF 2024Audio classification

    Transfer-learned MobileNet-v2 on spectrograms to identify under-studied Indian bird species.

  • Higgs Boson DetectionClassification

    Compared scikit-learn classifiers; 11th of 52 on the public leaderboard.

  • OpenAI CarRacingTensorFlow

    2D CNN driving agent in TensorFlow for randomly generated tracks.

  • Full Cycle TakeOutHACC 2023 finalist

    MERN and Meteor web app for ZeroWaste Oʻahu.

  • Protein foldingAlgorithms

    Hydrophobic-polar lattice model solved by recursive backtracking.

  • PANDA programming clubLecturer & Curriculum Developer · 2022–2025

    Designed an algorithms curriculum and led small-group sessions; 2023 Best Officer award.

  • ACM at MānoaServer Administrator & Webmaster · 2023–2024

    Maintained the chapter website and built internal tools for the leadership team.

Contact

Email is the quickest way to reach me: [email protected]. I'm also on GitHub, Hugging Face and LinkedIn.