Won Kyung Do (도원경, 都源京)

I recently graduated with a Ph.D. in Mechanical Engineering from Stanford University and now work at a stealth startup. During my Ph.D., I was advised by Prof. Monroe Kennedy III in the ARM Lab.

Before coming to Stanford, I received B.S. in Mechanical Engineering from Seoul National University, advised by Prof. Dongjun Lee. I was fortunate to work as a research intern at Naver Labs in 2018.

Email  /  CV  /  Scholar  /  LinkedIn  /  Github

Research

I'm interested in tactile sensing, dexterous manipulation, small object handling, control theory, optimization, and human-robot interaction. I believe improving dexterity of the robots is essential for enabling them to perform precise tasks such as manipulating small objects autonomously. Representative papers are highlighted.

Combines finite element analysis and deep learning to estimate dense stress tensors, deformation, and contact forces from optical tactile images, enabling dexterous manipulation of deformable objects.

Uses differentiable material point simulation and calibrated optical rendering to model large tactile-sensor deformations and transfer simulation-trained perception and manipulation to real sensors.

Developed a novel optical tactile sensor that can grasp multi-scale objects from flat surfaces (from 1mm basil seeds and small paperclips to items nearly 15mm).

Describes a method that uses optical tactile sensing for efficient detection and mapping of objects embedded in soft materials, demonstrated with quartz beads beneath polyethylene foam.

Presents a novel method that enhances 3D Gaussian Splatting with optical tactile sensors for more accurate object representation in robotics, integrating tactile data with monocular depth images and a new variance-weighted loss function for improved scene synthesis across various materials.

Introduces a tactile sensor-equipped gripper with DenseTact 2.0, enhancing precision and success in grasping small objects in cluttered environments, integrated with a specialized control algorithm and dataset for effective object classification and manipulation.

Designed DenseTact 2.0, an advanced optical-tactile sensor that accurately reconstructs shapes and measures forces in robotic fingertips with smaller form factors and higher resolution than previous models.

Introduces Densetact, a cost-effective, high-resolution tactile sensor that uses a fisheye camera and deep neural networks for precise 3D surface modeling in real-time.

Integrates sensor gloves and stereo cameras with a fusion algorithm for precise and reliable hand tracking, improving performance in real-world applications despite challenges like occlusions and electromagnetic interference.

Introduces a framework combining multi-modal tactile sensing and modular design with a classifier for recognizing social touch patterns, achieving up to 88.86% accuracy using methods like HMM, LSTM, and 3D-CNN.

Showed that the robot Shelly effectively reduces children's robot abusing while maintaining their engagement with the robot.

Designed "Shelly", a tortoise-like robot that engages with children while mitigating abusive behaviors towards robots.

Ph.D. Dissertation

Improving Robotic Dexterity with Optical Tactile Sensor DenseTact
Won Kyung Do
Ph.D. dissertation, Mechanical Engineering, Stanford University, 2025
dissertation

Patents

Optical Tactile Sensor
Monroe Kennedy III, Won Kyung Do
U.S. Patent Application No. 18/714,268
Publication No. US20250020523A1, 2025
published application

Service

Template taken from here.