WIN Lecture
Robotics Engineering Colloquium Series
Raven Huang
Object Search and Sensing
Wednesday, August 30th, 2023
12:00 PM - 1:00 PM
Unity Hall Rm 520
Abstract: Object search, especially for occluded objects and objects outside the view, is a challenging problem, yet a key step of many manipulation tasks. Geometry information can provide a clue of where the target object could be. Further, real life scenes are often semantically organized, meaning the surrounding objects are often semantically related to the target object. Large vision and language models (VLMs and LLMs) show promise for such relationships as they are pretrained on internet-scale data and capture semantic information. In this talk, I will discuss how this information, especially the semantic information, can be utilized to facilitate search using pre-trained VLMs and LLMs. Besides, additional modalities such as tactile may be needed for searching objects with specific textures. In this talk, I’ll also talk about how to build the connection between the vision and tactile and its preliminary applications in object search
Bio: Raven Huang received the B.S. degree from Shanghai Jiao Tong University and the M.S. degree from UT Austin. She is currently a Ph.D. candidate in the Department of Electrical Engineering and Computer Science, UC Berkeley. She has been a researcher with UC Berkeley’s AUTOLab since 2020. Her current research interests include robot learning, object retrieval, and locomotion.