Robot learning for precise, contact‑rich manipulation
Ph.D. Candidate in Computer Science · Rice University
0.02 mm zero-shot insertion20,480 simulated robots9 RSS/ICRA/IROS papers
I am a fifth-year Ph.D. candidate at Rice University, advised by Prof. Kaiyu Hang in the Rice Robot π Lab. I work on reinforcement learning and zero-shot sim-to-real transfer for contact-rich manipulation, and on robot self-calibration and self-identification, where robots learn about their own bodies, sensors, and workspaces by interacting with them.
I am looking for robot learning research internships and full-time research scientist / engineer roles after graduation (Fall 2027).
Before Rice, I received my B.S. in Computer Science from the University of Washington, where I worked on the MuSHR platform in the Personal Robotics Lab with Prof. Siddhartha S. Srinivasa. Outside of research, I enjoy cycling and bouldering.
Research Highlights
OmniState
First author · manuscript in preparation
Zero-shot sim-to-real peg insertion at a 0.02 mm tolerance
A reinforcement learning policy trained only in simulation, on 20,480 parallel Isaac Lab robots, inserts pegs on a real Franka at the 0.02 mm ManipulationNet tolerance, tighter than any it saw in training (per-shape hardware evaluation in progress). It commands full 6-D pose and impedance and senses contact only through the arm's joint torques. During training, simulated instances of the robot with different hidden board and grasp poses share one action, so the policy learns actions that work for every pose it cannot yet rule out.
UniARC-Calib and ARC-Calib
First author · IROS 2025 · T-RO submission
Markerless hand-eye calibration from the robot's own motion
The robot calibrates its cameras by itself: exploratory joint motions create image trajectories that a single geometric solver turns into the camera pose, for both eye-in-hand and eye-to-hand cameras, with no markers, CAD models, or learned detectors. On a Franka Research 3, eye-to-hand error is the lowest of 10 marker-based and markerless methods.
ARC-Calib paper videoIn-Hand Self-Identification and MPC
First author · IROS 2023
A robot hand learns its own control model from as few as 15 samples
A sensorless, underactuated Yale Model O hand identifies a local hand-object model online from as few as 15 exploratory samples, then follows reference trajectories with model predictive control, re-identifying itself when the contact dynamics change. No analytical model or offline training data needed.
arXiv IEEE video
ManipulationNet
Founding developer team · with NIST
A global benchmark for real-world robot manipulation
An infrastructure, launched with NIST, for benchmarking real-world manipulation skills and embodied multimodal reasoning across labs and robot platforms. I co-developed standardized physical-skill protocols for block arrangement and tabletop management.
website paper code NIST spotlightNews
- Oct 2026OmniState policies, trained only in simulation, insert pegs at the 0.02 mm tolerance on a real Franka, zero-shot.
- Sep 2026RoboTok, a data engine for retrieving internet demonstration videos for dexterous manipulation learning, is on arXiv.
- Jul 2026First-author journal paper UniARC-Calib submitted to IEEE Transactions on Robotics (T-RO).
- Jun 2026ManipulationNet featured in a NIST Spotlight.
- Jun 2026EgoInfinity, a web-scale 4D hand-object interaction data engine, is on arXiv.
- Mar 2026The ManipulationNet paper is on arXiv.
- Oct 2025ManipulationNet launched with NIST for real-world manipulation benchmarking at scale.
- Jun 2025Four papers accepted to IROS 2025, including first-author ARC-Calib.
- Apr 2025Robust peg-in-hole assembly accepted to RSS 2025.
Publications
2026
OmniState: A Shared-Action State Representation for Contact-Rich Manipulation under Uncertainty
UniARC-Calib: Unifying Autonomous Markerless Robot Calibration for Eye-in-Hand and Eye-to-Hand Setups
RoboTok: A Scalable Data Engine for Internet Demonstration Video Retrieval and Dexterous Manipulation Learning
EgoInfinity: A Web-Scale 4D Hand-Object Interaction Data Engine for Any-View Robot Retargeting and Video-to-Action Robot Learning
ManipulationNet: An Infrastructure for Benchmarking Real-World Robot Manipulation with Physical Skill Challenges and Embodied Multimodal Reasoning
2025
ARC-Calib: Autonomous Markerless Camera-to-Robot Calibration via Exploratory Robot Motions
Robust Peg-in-Hole Assembly under Uncertainties via Compliant and Interactive Contact-Rich Manipulation
rt-RISeg: Real-Time Model-Free Robot Interactive Segmentation for Active Instance-Level Object Understanding
On the Role of Jacobians in Robust Manipulation
Wearable Roller Rings to Enable Robot Dexterous In-Hand Manipulation through Active Surfaces
2024
Interactive Robot-Environment Self-Calibration via Compliant Exploratory Actions
Direct Self-Identification of Inverse Jacobians for Dexterous Manipulation Through Particle Filtering
2023
Non-Parametric Self-Identification and Model Predictive Control of Dexterous In-Hand Manipulation
Kinodynamic Rapidly-Exploring Random Forest for Rearrangement-Based Nonprehensile Manipulation
Experience
-
Graduate Student Researcher
Rice Robot π Lab @ Rice UniversityDeveloping reinforcement learning and zero-shot sim-to-real methods for contact-rich manipulation, and autonomous self-calibration and self-identification methods that let robots understand their own bodies, sensors, and workspaces. Co-developing the ManipulationNet benchmark and organizing the RGMC in-hand manipulation track.
-
Undergraduate Researcher, continuing after graduation
Personal Robotics Lab @ UWDeveloped a multi-agent coordination planner for MuSHR robots, integrating it with navigation systems for real-world task execution. Connected the iGibson and MuSHR simulators to enable perception-based navigation using real-world sensors and image processing.
-
Application Developer
StarbucksManaged, developed, and monitored data pipelines and integrations across Azure, Databricks, Incorta, Hadoop Hive, and Oracle Database for multiple teams.
-
Cloud Service Engineer Intern
StarbucksBuilt a Swagger API and integrated it with Cosmos DB and Azure AD, enabling efficient internal cloud queries. Developed a CI/CD pipeline to automate Python web app deployment to VMs behind a load balancer.
-
Software Engineer Intern
AgodaDeveloped R scripts that generated ad templates and summary messages for search engine advertising.
Service, Teaching, and Honors
Leadership and Service
- ManipulationNet, founding developer team (2025 – present): standardized physical-skill protocols for block arrangement and tabletop management.
- Robotic Grasping and Manipulation Competition, organizer of the in-hand manipulation subtrack, 9th and 10th RGMC (2024 – 2025).
- Thai Students Association, University of Washington, volunteer officer.
Teaching
- Teaching Assistant, Rice University: Senior Design in a Robotized World; Probabilistic Algorithms and Data Structures; Operating Systems and Concurrent Programming.
Honors
- Royal Thai Government Scholarship (2015): full tuition and stipend from B.S. through Ph.D.
- Competitive programming: ACM-ICPC UW qualifier, 1st-place team, and Pacific Northwest Regional, 16th place (2016); Thailand Olympiad in Informatics, Gold Medal (2014).
