Guowei Shi (石国维)

I'm a master student in the Department of Mechanical and Energy Engineering at Southern University of Science and Technology (SUSTech), under the supervision of Prof. Zhenzhong Jia. I graduated from SUSTech in 2022 with a Bachelor of Engineering in Robotics Engineering, where I was advised by Prof. Wei Zhang.

My current research focuses on contact perception and control for dynamic legged locomotion in challenging environments. My long-term goal is to develop robotic systems capable of exploring, understanding, and interacting with the physical world.

Email:12232336@mail.sustech.edu.cn  /  CV  /  Google Scholar  /  Github  /  Linkedin

Publications

(* indicates equal contribution, ** indicates corresponding authors)

We proposed a vision-based sensorized foot for dynamic legged robots, capable of simultaneously sensing contact forces and torques, estimating contact surface inclination, and perceiving foot-terrain interaction using only one camera.

We proposed an optimization method based on terrain mechanics to improve the studs of sensorized planar feet. Extensive experiments demonstrated that the optimized design significantly enhances the mobility of quadruped robots on granular slopes.

We designed a sensor suite for mobile robots, featuring RGBD cameras, Lidar, IMUs, and a global positioning system. Using this setup, we collected a comprehensive multi-modal SLAM dataset focused on robot locomotion over deformable granular terrains.

We designed a multi-modal sensorized planar foot for quadruped robots, equipped with IMUs, joint encoders, and force/torque sensors, enabling accurate estimation of local surface inclination and contact forces.

We analyzed foot-terrain interactions on deformable soils using sensorized feet, providing physics-informed insights to predict terrain properties and improve quadruped robot locomotion in challenging environments.

We presented the TAIL-Plus dataset and supplementary materials at the ICRA 2024 Workshop on Field Robotics, offering valuable resources for SLAM research in granular and challenging planetary-like environments.

We present a Wheeled-Bipedal Jumping Dynamics (W-JBD) model for optimizing height control, along with a Bayesian Optimization for Torque Planning (BOTP) method. This joint optimization framework enables precise height control and minimal energy consumption during jumping maneuvers.

Projects

We designed a full-sized humanoid robot with 20 degrees of freedom (DoFs) using quasi-direct drive (QDD) motors. The ongoing work involves implementing model predictive control (MPC), whole-body control (WBC), and reinforcement learning (RL) to achieve comprehensive whole-body motion planning.

We designed a 20kg quadruped robot with 12 degrees of freedom (DoFs) using quasi-direct drive (QDD) motors. test The ongoing work involves implementing model predictive control (MPC), whole-body control (WBC), and reinforcement learning (RL) to enable comprehensive whole-body motion planning.

Train reinforcement learning algorithm using CTS framework [Ref] and deploy on Unitree A1.

Implemented automatic image collection and cleaning using the MOG2 algorithm. Deployed and trained YOLOv5 for fast waste classification. Successfully planned and executed robotic arm motions to pick and toss waste onto a dynamic conveyor belt.

Academic Service

• Journal Reviewer: T-MECH, RA-L
• Conference Reviewer: ROBIO, ICARM