Wenliang Liu
Hello! I'm Wenliang Liu
I am currently a Postdoctoral Scientist at Amazon Robotics, where I work on planning and coordination for large-scale multi-robot systems. I received my Ph.D. from the Boston University Robotics Lab, advised by Prof. Calin Belta. My doctoral research lay at the intersection of machine learning, control theory, and formal methods, with a focus on learning-based control of robotic systems under temporal logic specifications.
SELECTED RESEARCH PROJECTS
Safe Learning under Signal Temporal Logic (STL) Specifications
We pioneered the utilization of Recurrent Neural Networks (RNN) as controllers to satisfy STL specifications on robotics systems, effectively addressing the historical dependency challenge associated with STL. We train the RNN using various machine learning techniques, such as imitation learning and model-based reinforcement learning. We also integrated Control Barrier Functions (CBF) into these algorithms to make both training and deploying safe.
Publications:
IEEE Control Systems Letters, presented in American Control Conference (ACC) 2021 (virtual)
IEEE International Conference on Robotics and Automation (ICRA) 2023, London, UK
IEEE Conference on Decision and Control (CDC) 2023, Singapore
Multi-agent Learning under Temporal Logic Specifications
We designed a novel temporal logic, called Capability Temporal Logic plus, or CaTL+ (pronounced cattle plus), which is specifically tailored for heterogeneous multi-agent systems. Then we developed a neural network framework to simultaneously learn the distributed control and communication policies, called CatlNet (pronounced cattle net), which has good scalability for large robotic teams.
Publications:
IEEE American Control Conference (ACC) 2023, San Diego, USA
5th Annual Learning for Dynamics & Control Conference (L4DC) 2023, Philadelphia, USA
Undergraduate Research
Vision-based robot grasping: Collected and labeled data from the camera on a Baxter robot’s hand to train a CNN for instance segmentation. Designed and implemented a system for a Baxter robot to grasp objects using visual information.
EDUCATION
2019-2024
Ph.D.
BOSTON UNIVERSITY
Boston, MA, USA
Mechanical Engineering
2015-2019
BEIHANG UNIVERSITY
Beijing, China
Instrumentation Science
Mathmatics
Bachelor
2012-2015
Senior High School
EXPERIMENTAL HIGH SCHOOL ATTACHED TO BEIJING NORMAL UNIVERSITY
Beijing, China
2009-2012
BEIJING NO.5 MIDDLE SCHOOL
Beijing, China
Junior High School