Myungsun Park (박명선)

Myungsun Park (박명선)

Postdoc at UC San Diego

Research

My research focuses on implementing human-like Physical AI in dexterous robot hands by integrating human hand motion and robot tactile sensing. My ultimate goal is to enable robots to perform complex tasks in uncertain, dynamic, and contact-rich environments through 1) precise motion, 2) sophisticated sensing, and 3) high-level intelligence.

Dexterous Motion Capture using Wearable Glove

Humans perform complex manipulation tasks through the coordination of multiple fingers. Implementing this level of dexterous robotic manipulation via Physical AI requires high-quality motion datasets, which are often challenging to acquire. I address this issue by developing wearable devices that accurately measure all degrees of freedom in human fingers. This approach enables intuitive teleoperation of dexterous robotic hands, facilitating complex tasks that are otherwise difficult to demonstrate.
[View details]
Hand Motion Capture and Post-processing with Enhanced Accuracy

Robust rotation-based visual motion capture outperformed position-based counterpart in accuracy. Geometry decoupling solved the retargeting problem by separating user-independent and dependent variables.

Wearable Motion Capture Glove

Calibration-free sensing achieved state-of-the-art accuracy using highly stretchable strain sensors that conform and auto-calibrate to diverse hand shapes. Versatile application demonstrated high performance in various virtual reality and human-robot interaction scenarios.

Intuitive Teleoperation of Multi-finger Robotic Hands

Seamless human-robot interface enables intuitive control of multi-DoF robot hands. Dexterous manipulation implemented precise fingertip control and multi-finger coordination for complex tasks.

Tactile Sensing and Haptic Feedback for Sophisticated Interaction

Humans perceive and interact with uncertain objects and environments using not only complex finger movements but also sophisticated tactile sensations. My research focuses on developing haptic feedback devices and tactile/environmental sensing technologies to implement this level of refined interaction in robots. This approach enables the development of multimodal tactile sensors for enhanced manipulation and adaptive robots in diverse environments.
[View details]
Haptic Feedback Devices for Teleoperation

IPMC actuator-based haptic feedback achieved enhanced remote grasping using a miniature gripper.

Benchmarking Tactile Sensors for Imitation Learning

Visuo-tactile policy evaluation identified the impact of six different tactile modalities on imitation learning for various manipulation tasks.

Development of Multimodal Sensors and Underwater Applications

Multi-modal flow sensors enabled simultaneous estimation of complex flow structures and robotic movement via shear and high-frequency components in fluid-structure interaction.

Human-Inspired Physical AI for Dexterous Manipulation

Human intelligence, characterized by semantic reasoning and purposeful action based on accurate motion and sophisticated sensing, can be implemented in robots. My research focuses on modeling human manipulation strategies and developing corresponding robotic sensing and control systems. Beyond mere demonstration or data dependency, I aim to realize efficient Physical AI for robots that mimics human-level decision-making and dexterity.
[View details]
Hybrid System Modeling for Vision-Free Manipulation

Defining hybrid variables modeled human semantic decision-making processes in grasping tasks. Proprioceptive sensing enabled complex manipulation without visual feedback.

State Estimation and Autonomous Gripper Systems

Proprioceptive sensor model guided the control system to robustly predict discrete physical events and contacts along with continuous poses of the robot and object.