Presenter: Weiyu Liu

Abstract: Robotics can transform our lives by providing assistance in various places, from homes to hospitals and disaster zones. Realizing this vision requires autonomous systems that can operate in unstructured environments, adapt to new situations, and interpret instructions from non-expert users. Yet current robots often fall short because they lack a general understanding of the world, the kind of common knowledge humans possess. In this talk, I will discuss frameworks driven by structured commonsense knowledge that addresses these challenges. We will focus on three key aspects: knowledge domains covering objects, spatial relations, and tasks; computational methods for jointly reasoning about language, vision, and physical actions; and knowledge sources including human-provided language instructions and large language models. By integrating these components, we move closer to developing robots that can effectively leverage prior knowledge, generalize, and provide meaningful assistance in real-world environments.

Bio: Weiyu Liu is a postdoctoral researcher in the Computer Science Department at Stanford University, working with Professor Jiajun Wu. His research brings together ideas from robotics, AI, and ML to develop robots with commonsense knowledge and the ability to operate autonomously in unstructured environments. Before joining Stanford, Weiyu completed his Ph.D. in Robotics at the Georgia Institute of Technology under the supervision of Professor Sonia Chernova. During his graduate studies, he interned at the NVIDIA Seattle Robotics Lab and collaborated with Facebook AI Research. He received his Bachelor’s degree in Electrical Engineering from Georgia Tech. He was selected as a Robotics: Science and Systems (RSS) Pioneer in 2022 and contributed to the 2024 US National Robotics Roadmap.

Hosted by: Professor Ian Lane and Professor Yi Zhang