Research

Research Vision

We develop embodied intelligence for companion robots that can move naturally, interact meaningfully with people, and manipulate the physical world. Our research combines robot learning, motion intelligence, multimodal foundation models, optimization, and human-robot interaction.

Research Topics

1. Natural Motion Intelligence

We develop learning- and optimization-based methods that enable humanoid and quadruped robots to reproduce, generate, and adapt natural motion across different embodiments.

STMR: Spatio-Temporal Motion Retargeting for Quadruped Robots (T-RO 2025, ICRA 2026)
RMR: Robust and Expressive Humanoid Motion Retargeting (IROS 2025)
CoRe: Contact-Aware Humanoid Motion Generation (Humanoids 2025)
Adaptive-HDM: Controllable Smoothness in Continuous Motion Generation (Computational Visual Media 2026)
CLIP-Actor-X: Text-Driven 4D Human Avatar Generation (TPAMI 2026)

2. Interactive Companion Robots

We study how robots can understand human intent, express dynamic personas, navigate socially, and engage people through playful and physical interaction.

MASK: Persona-Based Human-Robot Interaction Using AMBIDEX (RA-L 2024, ICRA 2025)
Turkish Ice Cream Robot: Playful Deception in Human-Robot Interaction (ICRA 2026)
PioneeR: Learning Social Navigation from Demonstrations and Rules (ICRA 2026)
Handover: Dynamic Robot-to-Human Handover (ICRA 2025)

3. Dexterous and Generalist Robots

We combine dexterous robotic hands, multimodal foundation models, perception, planning, and large-scale robot data to build generalist manipulation systems.

GRIT: Dexterous Grasping from Sparse Taxonomy Guidance (IROS 2026)
SoftHand: High-DOF Tendon-Driven Soft Hand (IROS 2025)
VIM: Unified Motion Reasoning and Generation (ICCV 2025)
@RLWRLD: Pouring Milk, Pick and Place, and Writing Down

Research Foundations

Our research is supported by multimodal foundation models, reinforcement learning, optimization, planning, perception, simulation, and robot data collection.

NFG: Natural Functional Gradients for Smooth Trajectory Optimization (RSS 2026)

TAD: Efficient Timing-Aware Planning via Detachable Task Modeling (IROS 2026)

VIUGIC: Transparent Object Pose Estimation via Depth Completion and Confidence-Guided Registration (IROS 2026)