Date: October 16, 2026
Speaker: Yaqi Xie, Assistant Professor, University of Illinois at Urbana-Champaign (UIUC)
Title: Compositional Generalization in Long-Horizon Robotics: From Understanding to Learning Structure
Abstract: Building trustworthy robot partners alongside humans faces a reality that scaling alone cannot fix: embodied data is orders of magnitude scarcer than language and vision data. Before sufficient data can be collected, agents need to make smart use of what we have by reasoning with structure. In this talk, I show how structure, from understanding it to learning it, enables robots to learn efficiently and generalize compositionally to long-horizon tasks. I first introduce how given structure, such as logic rules and knowledge graphs, can be embedded directly into neural architectures, integrating it into representations rather than attaching reasoning as a post-hoc module. I then demonstrate how such guided structure supports proactive human-robot collaboration, where robots reason over objects, parts, and affordances to interpret human intent, anticipate actions, and act within their own capabilities. However, predefined structure is static and incomplete. I therefore present how robots can learn the structure itself, inventing symbolic abstractions from experience, composing tools and skills, and growing a reusable skill library over time without forgetting. Together, these efforts outline a pathway toward robots that build, organize, and compose structure in new situations as genuine partners for humans.
Bio: Yaqi Xie is an Assistant Professor in Electrical and Computer Engineering and Computer Science at the University of Illinois Urbana-Champaign (UIUC). Before joining UIUC, she was a Postdoctoral Fellow at the Robotics Institute at Carnegie Mellon University (CMU), and she received her Ph.D. in Computer Science from the National University of Singapore. Her research develops trustworthy embodied agents, bridging foundation models and symbolic reasoning to build robot partners that are adaptive, interpretable, and safe alongside humans. She was named a Rising Star at NERC 2026 at Princeton University, and her work has been recognized with Best Paper Awards at the RSS 2025 GenAI-HRI and ICRA 2025 Nonverbal Cues workshops. Her other distinctions include the NUS Achievement Award and the Keppel Award of Excellence. She serves the research community as an Area Chair at NeurIPS and as an organizer of the ACM/IEEE International Conference on Human-Robot Interaction (HRI 2024) and the Learning on Graphs Conference 2026.
