EN | 陈冠廷

Phone: (+86) 13365983865

Email: cgtgetting@163.com

Chengdu, China

Hi, I am Guanting Chen, an undergraduate student in Automation (Outstanding Engineer Program), School of Electrical Engineering, Sichuan University.

My current interests focus on robotics, embodied AI, and visual perception.

Education

Sichuan University (Project 985 / Double First-Class), Automation, 2023.09 - Present
Weighted required-course GPA (first 5 semesters): 89.78
Rank (first 5 semesters): 12/170 (Top 7%)
Core courses: Automatic Control (93), Intro to Programming (94), Data Structures and Algorithms (92), Computer Organization and Design (96), Computer Networks and Communication (95)

Localization & Mapping (SLAM) and Target Recognition: Deployed FAST-LIVO2 SLAM with Mid360 LiDAR for mapping and high-precision localization.
Global Map Relocalization: Independently innovated a highly robust relocalization module by combining Teaser++ global registration with GICP local refinement. Compared to traditional RM methods (~10s for small initial deviations), this approach achieves millisecond-level localization even with large initial deviations and significant occlusion.

My Improved Relocation Demo

RL Training and Deployment: Introduced reinforcement learning algorithms to a custom 12-DOF robot system for locomotion control based on the HIMLoco framework. Completed policy training, simulation verification, and deployment process design, achieving Sim-to-Sim and Sim-to-Real transfer to improve policy robustness and adaptability.
Full-Stack Capability: Independently managed the entire pipeline: structural design participation, URDF/XML modification, parameter tuning in legged_gym, Sim2Sim (Isaac Gym to MuJoCo) and Sim2Real bridging, and deployment on Jetson Xavier edge devices.
Features: Supports LAN web remote control. Utilizes libtorch C++ inference engine to output target joint actions based on angular velocity.
Next Steps: Transplanting the Extreme-Parkour project to the custom quadruped for obstacle courses, and exploring MPC/VMC algorithms to meet diverse terrain requirements.

Physical Quadruped Robot Locomotion Demo

2025 China Robot Competition & RoboCup China Open (FIRA Small Size): National Third Prize.
27th China Robot and AI Competition: Sichuan First Prize and National Second Prize.

Simulation and Testing Environment: Built high-fidelity simulation scenarios and developed testing environments in Isaac Sim for sweeping and grasping tasks.
Physical Interaction Design: Constructed physically realistic mobile sweeping robot chassis and 6-DOF robotic arms. Designed physical interaction environments for “sweeping” and “grasping” tasks and formulated multi-dimensional quantitative evaluation metrics.
Skills Gained: Proficient in scene model building in Isaac Sim and ROS/ROS2 communication for action control.

Project Background: Addressed urban underground drainage pipe issues (roots, obstacles) by developing a specialized multimodal visual model for sewer defect detection.
Core Work: Fine-tuned the Qwen3-VL model combining SFT and GRPO schemes with multimodal input processing. Optimized RL objective functions with IoU and format rewards to significantly improve detection accuracy.
Research Stages:
- STAGE 1: Supervised Fine-Tuning (SFT) based on the QWEN3-VL foundation model.
- STAGE 2: Group Relative Policy Optimization (GRPO) training loop to achieve robust defect detection capabilities.

Software & Systems: Ubuntu/Linux, Python, C/C++, ROS2. Proficient in robot navigation, localization, robotic arm manipulation, and 2D/3D target recognition.
Simulation Platforms: Isaac Gym, Isaac Sim, Gazebo, MuJoCo.
Hardware & Deployment: Embedded development, PCB design, STM32/ESP32/ESP8266, Jetson Xavier edge deployment, hardware production.
Self-Evaluation: Rich experience in project reproduction, hands-on debugging, and rapid team integration. Actively focusing on the long-term research of quadruped/humanoid robot control, upper limb/manipulator control, dexterous hands, and embodied AI.