CMU · M.S. AI Engineering · Robotics & Embodied AI

Researching and building robots that act in the physical world.

Research depth · GVLA, ECCV 2026 (2nd author); studying VLA generalization and decision-making.

Real-world robotics · Franka / UR5 evaluation and 50+ SO-101 demonstration episodes.

Engineering · Control, embedded hardware, simulation, and full-stack software.

I'm Zihong Luo (Logan), an M.S. AI Engineering student at Carnegie Mellon University working across robotics, embodied AI, and machine learning systems: from what VLM and VLA representations encode, to the controllers and robots that act on them.

The whole system as one stack: four layers, 13 parts.· scroll to take it apart

04Research Loop03Control & Simulation02Perception & Learning01Robot Systems
02 / 05

Research Focus

ONGOING · CMU

How do pretrained vision-language models encode space and physics, and does that knowledge reach the actions a robot takes? My work follows one loop:

  1. 01Observe

    Scenes, language, proprioception

    Camera views, an instruction, and the robot state. In LIBERO I build counterfactual scene pairs that differ in one geometric factor.

  2. 02Represent

    VLM / VLA features

    Pretrained VLMs such as Qwen3-VL produce visual-language features. The open question: do they encode spatial and physical structure, and does it survive into a VLA policy? Measured with representation probes.

  3. 03Act

    Action generation

    A VLA maps those features to robot actions. In GVLA the policy is also conditioned on gripper morphology, so one model can adapt its strategy to the end-effector.

  4. 04Validate

    Causal and real-world tests

    Activation interventions test whether a representation actually changes the action; rollouts in simulation and on Franka / UR5 measure task success.

The CMU work is in progress: these are the questions and methods, not yet findings.

03 / 05

Selected Publications

* EQUAL CONTRIBUTION

Selected for relevance to my current research. See the complete record on Google Scholar.

Under review

2027
Contributing Author
ICRA 2027Under Review

Graph-Guided Gripper-Aware Vision-Language-Action Models

..., Zihong Luo, et al.

04 / 05

About

Zihong Luo

Lead Student Researcher with Prof. Guannan Qu at CMU. Before that, research assistant at SmartLab, University of Liverpool, with Prof. Anh Nguyen, and BSc in Computing Science (AI) with First Class Honours. The question connecting my recent work is how multimodal robot policies can adapt to different grippers, tasks, and real-world execution constraints.

Recent

Research Experience

Graduate Researcher - Lead Student Researcher

Carnegie Mellon University

Pittsburgh, PA, USA · Advisor: Prof. Guannan Qu

Sep 2026 – Present
  • Leading research on how pretrained Vision-Language Models (VLMs) encode spatial and physical knowledge, and whether these representations transfer to Vision-Language-Action (VLA) policies for robotic manipulation.
  • Designing controlled experiments in LIBERO using Qwen3-VL and VLA models, with representation probing, counterfactual scene pairs, and activation interventions to investigate how geometric information influences downstream action generation.

Research Assistant

SmartLab, University of Liverpool

Liverpool, UK · Advisor: Prof. Anh Nguyen

Oct 2025 – Aug 2026
  • Core contributor to GVLA (ECCV 2026, 2nd author), a VLA that conditions manipulation on gripper morphology: co-designed the gripper-conditioning module from inception, implemented and ablated multiple architectural variants, ran experiments, and co-wrote the paper.
  • Worked across simulation and real hardware on multi-gripper Franka / UR5 platforms with dual RGB-D cameras: teleoperation, data collection, behavior cloning, and sim-to-real evaluation.
  • GCA-Bench (IROS 2026): complex-grasping benchmark from detection to execution; contributed real-robot data collection, dataset organization, and manuscript preparation.

Software Engineering Contributor

Donkey (Y Combinator S26)

Remote

Sep 2025 – Jan 2026
  • Contributed to full-stack development (frontend, backend, databases) and AI-agent orchestration for an AI-native sourcing platform (donkey.trade).

Algorithm Intern

Jifu Medical (AI Algorithm Group)

Shenzhen, China

Jun 2025 – Aug 2025
  • Context: Jifu Medical makes magnetically controlled capsule endoscopy. A patient swallows a capsule camera, and an external magnet on a robotic arm steers it through the stomach and intestine, so a full examination needs no anaesthesia. That matters for elderly patients who cannot be sedated, and it spares patients the discomfort of a conventional scope.
  • Doctors drove the magnet arm by hand. I joined to help build the first VLA-style autonomous arm system: an external camera estimates the patient's posture, the capsule's own images show which regions have been covered, the system judges whether the whole stomach has been scanned, and the arm adjusts the capsule's position, direction and orientation on its own.
  • Built the imitation-learning groundwork on SO-101 dual arms with ACT, ALOHA and LeRobot: teleoperation, synchronized multi-view recording, and joint-state / action logging; collected and validated 50+ real-robot episodes.

Research Assistant

MBZUAI

Abu Dhabi, UAE · Advisor: Prof. Imran Razzak

Dec 2024 – Jul 2025
  • Developed PG-SAM, a prior-guided medical image segmentation framework using medical LLMs to provide modality-aware semantic priors.
  • Designed a Modality Prior Aligner and fusion decoder connecting semantic reasoning with dense prediction. Outcome: Accepted to BIBM 2025.

Summer Research Assistant

University of Exeter

Exeter, UK · Advisor: Prof. Yanda Meng

Mar 2024 – Aug 2024
  • Contributed to IMDR, a multimodal representation learning framework for ophthalmic disease grading under incomplete and noisy modality settings.
  • Designed ablation studies and incomplete-modality evaluations. Outcome: Accepted at AAAI 2025.

Research Contributor

Tongji University School of Medicine

Shanghai, China · Advisor: Prof. Xiaoyun Xie

Nov 2023 – Jan 2024
  • Built interpretable ML models for DPN/LEAD prediction; applied SHAP for risk factor analysis aiding DFU prevention.

Research Assistant

XJTLU

Suzhou, China

Sep 2023 – Nov 2023
  • Developed encoder-decoder with Deep Belief Network for modality completion (ICPR 2024).
  • Integrated image + temporal signals via spiking networks for anomaly detection (ICPR 2024).

Education

Carnegie Mellon University

M.S. in AI Engineering - ECE

Pittsburgh, PA, USA

Aug 2026 - Dec 2027 (Expected)

University of Liverpool (UoL)

BSc in Computing Science (AI) | GPA: 3.9/4.0 | First Class Honours

Liverpool, UK

Sep 2022 - Jun 2026
05 / 05

Open to research collaboration and Summer 2027 internships in robot learning, VLA models, and robotics systems.