Thanks for stopping by. I‘m looking for a PHD position in (wheel-)legged robots.If you are interested in me, feel free to contact me!
I am a MPhil student, who studying in The Hong Kong University of Science and Technology, Guangzhou Campus, Robotics and Autonomous Systems.
☢ Github
📱 (86) 13518393890
👨🏫 Supervisors: **Ming LIU** ✉️ [email protected]
[**Jun MA](<https://scholar.google.com.sg/citations?hl=zh-CN&user=8VepsVAAAAAJ&view_op=list_works&sortby=pubdate>)** ✉️ [email protected]
Arm-Constrained Curriculum Learning for Loco-Manipulation of the Wheel-Legged Robot ( IROS Oral 2024) [Project Page][Github]
Contrastive Imitation Learning for Language-guided Multi-Task Robotic Manipulation (CORL 2024)[Project Page]
Mitigating the Human-Robot Domain Discrepancy in Visual Pre-training for Robotic Manipulation (Submitted to ICRA 2024)
MIP-MPC Framework for Wheeled-Legged Robot working on Tough Terrain(A gait -schedule-free MPC framework for legged and wheel-legged robots is undergoing physical and simulation experiments.)
Poster presentation in Autonomous Robotic Technology Seminar(ARTS 2023)
National Second Prize | Equestrian Simulation of Asia-Pacific Broadcasting Union 2020
• Contributed to the development of a segment of the Simulink simulation system and engaged in co-simulation using Adams, facilitating the execution of multiple maneuvers by the quadruped robot in a simulated environment.
National Third Prize | RoboMaster of DJI 2020
• Contributed to the development of the fundamental control framework utilizing Linear Quadratic Regulator (LQR) techniques. Implemented sensing algorithms for image processing and LiDAR data interpretation, and executed sensor data fusion. Additionally, contributed to the design of the vehicle’s hardware circuitry, encompassing aspects such as driver circuits and power supply systems.
National Second Prize | Asia-Pacific Broadcasting Union Robocon 2019
• Responsible for the leg motion control of the quadruped robot, the ejection and gripping device control of the upper structure of the four-wheel omnidirectional robot.
Provincial Second Prize | National College Student Engineering Training Competition 2020
• Responsible for hardware circuit design, fusion of the data of the orthogonal code disk and DT35 laser ranging module, estimation of the current position of the car body, and the communication protocol between different functional modules.
Introduction: This was the main research project I undertook during my postgraduate studies. We modified a quadruped robot into a wheel-legged robot, with the aim of combining the advantages of legged and wheeled robots, enabling it to operate in various dynamic environments. At present, we have completed the construction of the hardware platform and replicated the relevant papers from ETH. Building on this, in order to meet the need for flexible gait switching of wheeled robots, we proposed MIP-MPC to solve this problem and rewrote all gait-related constraints. We have already completed preliminary result verification in a simulation environment. Below are some demonstrations of the physical robot and experiments. The specific details on how to incorporate gait information into MPC and its corresponding constraints, and how to find the optimal solution, will be elaborated in our upcoming paper to be submitted to IEEE Robotics and Automation Letters (RAL). At the current stage, we are conducting simulation experiments and physical experiments.
Motion Planning and Control of Wheel-legged Robots
The robot’s structure
This is our wheeled robot, modified from the Unitree GO1. The structure of the robot’s wheels, the motor drive circuit, and the software were all designed by us. It is equipped with computing devices such as Nuc 13 and Orin Nx, as well as sensor designs like the Livox Mid360 LIDAR depth camera.
This work was completed between January and April 2023.
Automatically Switch Gait According to the Environmental
a demonstration of automatic gait transition. The left figure shows a visualization of the simulation data. The semi-transparent map is used to evaluate the terrain traversability. The right figure is a view of the physical simulation. The bottom right figure shows the real-time gait optimization results.
The robot is capable of generating a Risk Map based on its surrounding environment. We incorporate gait information as an optimization variable into the Model Predictive Control (MPC) for optimization. To integrate gait information with perception information, we linearize the Risk Map at specific points, thereby representing the perception information as a linear function of the robot’s end effector position. This allows for automatic gait transition and maintains the robot’s stable state without any given reference gait or reference foot trajectory.
This work continues up to the present.