Projects
You can find more details about each project on its website by clicking the title of the corresponding project.
Logic Learning from Demonstrations for Multi-step Manipulation Tasks in Dynamic Environments
This project aims to design a reactive task and motion planning algorithm with one single demonstration for long-horizon manipulation tasks in dynamics tasks. It also results in an LfD that can imitate, generalize, and reactive to disturbances for multi-step tasks in real world. We tested the method in several tasks and are targeting on to extent it for building intelligent robot assistants at home.
Representing Robot Geometry as Distance Fields: Applications to Whole-body Manipulation
This project represents a serial robot arms with Signed Distance Functions (SDF) with kinematic chain awareness. This differentiaable implicit representation enables efficient minimal distance and gradient query, which thus facilitates the design of dual-arm self-collision avoidance and whole-arm lifting big and bulky objects. I am mainly invovled in the dual-arm self-collision section.
Learning and Generalizing Variable Impedance Manipulation Skills from Human Demonstrations
This project aims to enable robot arms learn variable impedance manipulation skills from multiple demonstrations for pouring tasks in human-centric environments.
This project aims to reproduce the results presented in the paper titled Learning Agile Robotic Locomotion Skills by Imitating Animals on our self-designed multi-modal quadruped robot. I was involved in this project when I was doing my research internship at Tencent Robotics X Lab and my work mainly focus on the Sim2Real part. During my internship, I managed to transfer gaits learned in simulation to the real quadruped robot with a 100% success rate.
Learning Optimal Manipulation Skills in a Human-like Way for Contact-rich Tasks
This project targets on learning optimal variable impedance manipulation skills with human demonstrations in a residual reinforcement learning manner for contact-rich tasks, e.g. peg insertion. I managed to deploy it on a 7-axis Franka Emika robot arm by integrating deep reinforcement learning and imitation learning with Python, C++, and ROS.