Robotics Engineering Colloquium Series
Dr. Wanxin Jin, University of Pennsylvania
Learning and Control for Interactive Autonomy
Wednesday, February 22nd, 2023
1:30 PM - 2:30 PM
Unity Hall | Room 520
Abstract: From improving the quality of life for the aging population to agile manufacturing, robots are required to have interactive autonomy --- the capability of operating in cohorts with humans and skillfully manipulating physical objects. Existing techniques, built upon big data and deep models, are unable or inefficient to solve two critical challenges in interactive autonomy. The first is individualizability, which requires a robot to quickly understand user intent and adapt its behavior to user preference. The second is multi-modality, which requires a robot to make hybrid decisions to touch and affect physical objects for maximal dexterity. This talk aims to address the above two challenges by developing learning and control methods from the most accessible and efficient interactive data.
In this talk, the first part will focus on individualizability, where we will propose the methods that enable a robot to learn human intent from a snippet of human motion and adapt its autonomy toward human preference through intuitive human interactions. The second part will focus on multi-modality, where we will develop a technique that enables efficient learning of hybrid representation (thousands of modes) for contact-rich manipulation and then propose a task-driven hybrid model reduction method that, without any prior knowledge, solves the robotic in-hand dexterous manipulation within fewer minutes of online learning. With the goal of unifying techniques in the first two parts, the final part of this talk will introduce a general-purpose computational framework. By bridging tools in the control field and those in machine learning, this framework can solve a broad class of robot learning and control tasks while attaining higher efficiency and provable safety.
Bio: Wanxin Jin is a postdoctoral researcher in the GRASP Laboratory at the University of Pennsylvania. He received his Ph.D. degree in autonomy & control from the School of Aeronautics and Astronautics, Purdue University, in 2021. Wanxin’s research interests include robotics, control, and machine learning, with an emphasis on the autonomy of robots as they physically interact with humans and objects. Wanxin Jin is a recipient of the Purdue ICON Outstanding Research Awards, Purdue Magoon Award for Excellence in Teaching, Purdue Ross Fellowship, and Best Student Paper Finalist at IEEE DASC 2021. From 2016 to 2017, Wanxin Jin was a research assistant at the Technical University of Munich, Germany, and he holds B.E. and M.Sc. degrees from Harbin Institute of Technology, China.