Date: January 28, 2026
Speaker: Mingyo Seo, Ph.D. student, The University of Texas at Austin
Title: Embodiment-Aware Skill Learning for Diverse Robots
Abstract: Robotic systems are rapidly diversifying in both form and application, spanning humanoids, mobile manipulators, and industrial platforms. Achieving generalist robot autonomy therefore requires shareable skills that remain effective across a wide range of robots. Recent efforts have pursued this goal by scaling data across multiple robots, implicitly learning structural differences and invariances from large datasets. However, robots are engineered systems whose behavior is fundamentally shaped by hardware, control, and representation. Rather than expecting learning algorithms to rediscover this structure from data, I develop embodiment-aware skill learning, which incorporates these elements as structural priors to guide learning toward physically feasible and transferable behaviors.
In this talk, I will present my work on embodiment-aware skill learning through a holistic approach that integrates structural priors from robot systems into data-driven methods. I begin with hardware design for consistent physical interaction and sensing across diverse robot embodiments. I then introduce control abstractions that shape the learning problem within compact state–action spaces for complex robots. Finally, I discuss embodiment-aware representations that separate object-centric task intent from geometric constraints, addressing morphological differences in reachable states and feasible trajectories and enabling efficient skill transfer and robot-specific feasible behaviors.
Bio: Mingyo Seo is a Ph.D. student at The University of Texas at Austin, advised by Prof. Yuke Zhu and Prof. Luis Sentis. His research goal is to build generalist robot autonomy by enabling knowledge sharing and composition across diverse robot embodiments. His research takes a holistic, full-stack approach that integrates domain priors from robot systems into data-driven methods for embodiment-aware skill learning, improving sample efficiency and generalization. He received an M.S. from the University of Illinois at Urbana–Champaign and a B.Eng. from Tohoku University, and has spent time at the Boston Dynamics AI Institute (now the RAI Institute) and Dexterity.