RHOS: Robot, Human, Object, and Scene

Talk by Yong-lu LI

Oct 12, 2023 Thursday


Intelligent robot matters. Our research goal is to build a “brain” that enables agents to perceive human activities, reason human behavior logic, learn skills from human activities, and interact with the environment. In this talk, we will present our old and new works covering the understanding of human-object-scene, visual reasoning for human actions and object concepts, and robot learning for object manipulation. Our framework consists of three parts: 1. Grounding from the physical world to the symbolic concept space; 2. Casual reasoning based on extracted knowledge and concepts; 3. Agent skill learning and planning. Besides, we will discuss possible paths to future work on embodied and data-centric AI, given the recent dramatic changes in the AI landscape.


Oct 12, 2023 Thursday



RmE1-101, GZ Campus


628 334 1826 (PW: 234567)

Speaker Bio:

Yong-Lu Li

Assistant Professor, Shanghai Jiao Tong University

Yong-Lu Li is a tenure-track assistant professor at SJTU. In 2021-2022, he worked closely with IEEE fellow Prof. Chi Keung Tang and Prof. Yu-Wing Tai at HKUST as a postdoc fellow. He received a Ph.D. degree (2017-2021) in Computer Science from SJTU, under the supervision of Prof. Cewu Lu. His primary research interests are human activity understanding, visual reasoning, and embodied AI. His research projects include open-sourced systems HAKE (http://hake-mvig.cn, 115k pageviews) and AlphaPose (Github star 7.3k). 

His works were published in top-tier CV/AI/ML conferences and journals, such as TPAMI, NeurIPS, CVPR, ICCV, ECCV, etc. He has won Wu Wenjun Outstanding Doctoral Thesis Award, WAIC Yunfan Award – Shining Star (Global 10) & Rising Star (Global 10), Baidu Scholarship (Global 10), NeurIPS ’21/22 Outstanding Reviewer, Chinese Student AI 100 (ML, Global 10), Shanghai Outstanding graduates, Yang Yuanqing Education Fund – outstanding PhD.

Homepage: https://dirtyharrylyl.github.io/ 

Lab Website: https://mvig-rhos.com/