Perception — L4 robocar

• Developed intraoperative X-ray image calibration/distortion-correction and a PyTorch-based CT-to-X-ray 2D–3D registration pipeline for spine surgery, achieving 98%+ bead detection, 100% 3D–2D registration, and < 0.5-pixel reprojection error.

• Built multi-threaded C++ real-time control software for a medical laser (~100 ms response) with a Win32 UI, and a 3D–3D head registration/deformation workflow that performs sensor fusion on depth/ToF data from cameras (RealSense D405, OAK-D) using point-to-plane ICP and DBSCAN for image-guided brain surgery.

• Prototyped a CLIP-based vision–language pipeline for non-contact sleep monitoring, automatically analyzing posture and movement patterns during sleep.

• Collaborated with mechanical, systems, and electrical engineers, using Jira and CI/CD pipelines within an FDA-compliant QMS aligned with IEC 60601 and related medical device safety standards.

Two internships at KCL Robotics (May 2023 – Aug 2023, Jan 2024 – May 2024), working on ROS2-based perception pipelines and embedded software for an ultrasound image–guided needle injection medical device.

• Designed and implemented a ROS2 C++ integration pipeline with 14+ nodes/topics and launch files, coordinating two stereo cameras, a high-resolution RGB camera, and two thermal cameras for synchronized acquisition.

• Developed calibration and 3D perception modules, including multi-camera calibration, triangulation, and 3D reconstruction to support accurate tool localization.

• Implemented event-driven embedded firmware in C for a Cypress PSoC microcontroller on the ultrasound image–guided needle injection device, handling GPIO events, ADC sampling, safety checks, and communication with the host system.

• Contributed to a companion Android front-end app, enabling real-time visualization of device status and sensor data to support clinicians and service engineers.