Mingyo Seo
Papers
5
Total Citations
82
H-Index
4
About
Mingyo Seo is a roboticist pushing the boundaries of humanoid and quadrupedal locomotion and manipulation through deep imitation learning and model predictive control. Their key research areas include cross-embodiment imitation, perceptive locomotion in dynamic environments, and real-time task control for industrial manipulators. Seo’s most impactful work, "Deep Imitation Learning for Humanoid Loco-manipulation Through Human Teleoperation" (48 citations), introduces TRILL—a data-efficient framework that enables humanoids to learn complex loco-manipulation skills from human demonstrations, addressing the high-dimensional control challenge. They also developed PRELUDE, a hierarchical learning framework for quadrupedal robots to navigate cluttered and moving obstacles (9 citations), and LEGATO (6 citations), which enables cross-embodiment imitation using a grasping tool, significantly reducing the cost of policy transfer across different robots. Seo’s real-time MPC strategy for industrial manipulators (17 citations) ensures singularity-tolerant hierarchical task control, vital for safe, multi-task automation. Their recent work, OKAMI, teaches humanoid robots manipulation skills from a single video demonstration, showcasing object-aware retargeting. With over 80 total citations and a growing portfolio of innovative frameworks, Seo is shaping the future of agile, adaptable robots capable of learning from minimal human input.
Research Focus
Key Achievements
Top Papers
- 1
- 2
- 3
- 4LEGATO: Cross-Embodiment Imitation Using a Grasping Tool6 citations · 2025
- 5