CV
Download my CV as PDF
Education
- Carnegie Mellon University, Pittsburgh, USA
- M.S. in Mechanical Engineering - Research (Robotics Track), Aug. 2024 – May. 2026
- GPA: 4.0/4.0
- Dalian University of Technology, Dalian, China
- B.E. in Mechanical Design & Manufacturing and Their Automation (Japanese Intensive), Sept. 2019 – Jul. 2024
- GPA: 3.7/4.0
- Tohoku University, Sendai, Japan
- Exchange Student in Mechanical and Aerospace Engineering Department, Oct. 2022 – Aug. 2023
Publications
C=Conference, J=Journal, W=Workshop, A=arXiv
[A.1] LV-DOT: LiDAR-visual dynamic obstacle detection and tracking for autonomous robot navigation.
Zhefan Xu, Haoyu Shen, Xinming Han, Hanyu Jin, Kanlong Ye, Kenji Shimada
arXiv:2502.20607[W.1] Adaptive Planning Framework for UAV-Based Surface Inspection in Partially Unknown Indoor Environments.
Hanyu Jin, Zhefan Xu, Haoyu Shen, Xinming Han, Kanlong Ye, Kenji Shimada
ICRA 2025 Construction Robotics Workshop
Experience
Research Assistant - CERLAB, Carnegie Mellon University
Aug. 2024 – Present, Pittsburgh, USA
Supervisor: Kenji Shimada
- Implemented LSTM-PPO architecture for wind-resilient UAV RL-control, modeling various wind fields in Gazebo/Isaac Sim and training distributed wind-aware policies that improved UAV navigation robustness.
- Contributed to the development of the popular open project CERLAB UAV Autonomy stack in ROS, Gazebo, and Isaac Sim. Integrated point cloud reconstruction after dynamic obstacle removal.
- Built a custom LiDAR-based UAV platform and conducted real tunnel inspection tests for Toprise Inc Japan., achieving high-resolution 3D reconstruction (accuracy < 5cm).
Software Engineer Intern - Perflection AI
May. 2025 – Aug. 2025, Pittsburgh, USA
- Benchmarked state-of-the-art vision-language models (Qwen2.5-VL, Tarsier, GPT-4o, Gemini 2.5 Pro) for golf advice accuracy and curated domain-specific datasets for fine-tuning.
- Designed and iteratively refined prompt engineering strategies, transforming generic outputs into personalized and actionable coaching feedback.
- Incorporated key frame analysis and biomechanical metrics into the feedback pipeline, improving swing issue detection and user trust in model-generated advice.
Projects
For detailed project descriptions, please visit the Projects page.
- Pittsburgh-RAG: Retrieval-Augmented Generation for QA on Pittsburgh (Sep. 2025 - Oct. 2025)
- Build-LLAMA2: Mini Llama2 Transformer Implementation (Aug. 2025 - Sep. 2025)
- DDPM-AFHQ: Denoising Diffusion Probabilistic Model on AFHQ Dataset (Jun. 2025 - Aug. 2025)
- Windy-NavRL: Wind-resilient RL Framework for UAV Navigation (Feb. 2025 - Present)
- ORB-SLAM3 on Various Physical Robots using ROS2 (Sep. 2024 - Dec. 2024)
- Optimal Control and A* Path Planning for Autonomous Vehicles (Aug. 2024 - Dec. 2024)
Skills
- Languages: Chinese (Native), English (Fluent), Japanese (Fluent)
- Programming: C/C++, Python, MATLAB, Git, JavaScript
- Frameworks & Libraries: PyTorch, NumPy, OpenCV, ROS
- Robotics & AI: Path Planning, SLAM, Object Detection, RL, LLM, VLM
- Software & Tools: Isaac-sim, Gazebo, MuJuCo, AutoCAD, SolidWorks, Ansys, Wandb
