Experiences

Robotics Perception and Manipulation Lab | University of Minnesota

Researching data-efficient robot learning and manipulation in Prof. Karthik Desingh's RPM Lab. Led the design and implementation of a goal-conditioned positioning policy for category-level last-meter navigation, using multi-view RGB observations collected on Boston Dynamics Spot, resulting in an RA-L submission.

Exploring one-shot generalization for eye-in-hand manipulation via in-context imitation learning from a single wrist-mounted demonstration. Also built SpotStack, a modular software library for perception integration, data collection, and policy deployment on Spot.

CSCI 5551: Introduction to Intelligent Robotic Systems | University of Minnesota

Designed and built a 10-stage real-robot manipulation curriculum using a 6-DoF Lite6 robotic arm and ZED 2i stereo camera. The curriculum progresses from AprilTag-guided manipulation to pure vision-based manipulation, covering single-object grasping, target selection, and multi-object sequential stacking.

Built reusable software infrastructure including AprilTag-based camera-to-robot calibration, grasp primitives, and camera wrapper APIs. Implemented reference solutions using SAM-based segmentation and oriented bounding box fitting on point clouds for zero-shot target selection and 6D object pose estimation.

CSCI 5561: Computer Vision | University of Minnesota

Designed and maintained an automated grading system for coding assignments, enabling scalable and consistent evaluation across student submissions. Held weekly office hours and provided technical guidance on computer vision algorithms and deep learning methods, including feature extraction, image registration, scene recognition, and CNN implementation.

Contributed to midterm exam design and grading, and managed overall course grading logistics. Received strong student feedback on teaching clarity and approachability, as reflected in the official SRT results.

Industrial Technology Research Institute (ITRI) | Taiwan

Established an autonomous grinding production line for irregularly shaped water faucets using a robot arm and a force feedback sensor. Designed force-adaptive trajectories that maintain consistent contact force throughout the grinding process, enabling uniform surface finishing across the complex and irregular geometry of each faucet.

Improved on-site calibration efficiency by designing a coordinate transformation algorithm for adaptive trajectory tuning, making it easier for operators to recalibrate the system between production runs.

NYCU Autonomous Underwater Vehicle (AUV) Team | Taiwan

Led the design and implementation of the software architecture, control system, and navigation for the team's AUV. Established a modular high-level/low-level system using an STM32 embedded controller for real-time attitude and motion control, with ROS deployed on Raspberry Pi and Jetson Nano for task planning and navigation.

Reduced high-frequency noise in attitude estimation by 10% by implementing a gradient descent-based Madgwick filter on embedded hardware. Developed a real-time obstacle-aware navigation system using the A* algorithm with sonar-based localization.