Skip to content
View JialiangFan's full-sized avatar
🎯
Focusing
🎯
Focusing

Block or report JialiangFan

Report abuse

Contact GitHub support about this user’s behavior. Learn more about reporting abuse.

Report abuse
JialiangFan/README.md

Jialiang Fan (樊佳亮)

PhD student in Computer Science & Engineering at the University of Notre Dame, advised by Prof. Fanxin Kong. I work on the safety of embodied AI — making robot-learning and LLM-driven autonomy systems behave safely and verifiably in the real world.

My research sits at the intersection of vision-language-action (VLA) models, safe reinforcement learning, formal methods, and cyber-physical systems: I build benchmarks and methods that expose where "successful" robot policies are actually unsafe, and design neural-symbolic and formal-method-guided approaches to close that gap.

Research Interests

  • Safe & reliable embodied AI / vision-language-action (VLA) models
  • Safe reinforcement learning and its vulnerabilities
  • Cyber-physical systems and formal methods (STL, neural-symbolic verification)
  • LLM-based task planning for robotic systems
  • World models and robot learning

Selected Publications (2026)

  • SafeVLA-Bench: A Benchmark for the Success–Safety Gap in Vision-Language-Action ModelsarXiv 2026 (first author). arXiv
  • SafeGen-LLM: Enhancing Safety Generalization in Task Planning for Robotic SystemsarXiv 2026 (first author). arXiv
  • Vulnerability Analysis of Safe Reinforcement Learning via Inverse Constrained Reinforcement LearningarXiv 2026 (first author). arXiv
  • Vulnerability Exploration of Safe Reinforcement Learning in Cyber-Physical Systems via STL MiningICCPS 2026 (first author).
  • SafeNet: A Neural-Symbolic Network for Safe Planning in Robotic Systems using Formal Method-Guided LLM Fine-TuningICRA 2026 (co-author).

Full list & citations on Google Scholar.

Selected Projects

  • awesome-vla-benchmarks — a curated list of benchmarks for Vision-Language-Action models in robotics.
  • vla_mycobot — run VLA / imitation-learning policies on a myCobot 280Pi (6-DOF arm + RPi 4): LeRobot adapter, teleop, ACT training pipeline.
  • research-project-guide — conventions for organizing reproducible research projects (docs/, code/, runs/).

Education

  • Ph.D., Computer Science & Engineering — University of Notre Dame (2024–present)
  • M.S., Computer Technology — Lanzhou University (2023)
  • B.E., Software Engineering — Shandong University (2019)

Experience & Awards

  • Research Intern, ABB (Summer 2026)
  • Outstanding Teaching Assistant Award, Notre Dame CSE (2025–2026)

Links

Pinned Loading

  1. awesome-vla-benchmarks awesome-vla-benchmarks Public

    A curated list of benchmarks for Vision-Language-Action (VLA) models in robotics

    4

  2. vla_mycobot vla_mycobot Public

    Run VLA / imitation-learning policies on a myCobot 280Pi (6-DOF arm + RPi 4). LeRobot adapter, teleop, ACT training pipeline.

    Python