Danfei Xu
Date:
Speaker
Danfei Xu is an Assistant Professor in the School of Interactive Computing at Georgia Institute of Technology. He received his Ph.D. in Computer Science from Stanford University in 2021. His research is in machine learning methods for robotics, with a focus on manipulation planning and imitation learning. His research goal is to enable physical autonomy in everyday human environments with minimum expert intervention.
Speakers Links: Google Scholar - Website,
Abstract
From making a cup of coffee to navigating a foreign city, humans are remarkable at devising plans and orchestrating actions to achieve their goals. How can we build robots that can do the same? This talk will present a view rooted in the hierarchical decision making paradigm that emulates reasoning and acting computationally. I will discuss why the classical view of modeling acting and reasoning in silos is problematic, and how learning from experience can help bridging the two and enabling new capabilities. Specifically, I will describe two works that attempt to make the connection in each direction — from reasoning to acting and vice versa. The first work learns manipulation skills by leveraging the reasoning abstraction of a TAMP system. The second work builds skill-level world models for long-horizon planning. If time allows, I will conclude the talk by discussing how we might scale to true everyday autonomy through addressing data problems at different granularity levels and making use of foundation models.