cv

BASc in Engineering Physics

University of British Columbia

4th year, specializing in robotics and autonomous systems

• Machine learning including supervised and unsupervised learning, VAE, neural networks, and reinforcement learning.
• Computer vision including feature detection and matching, stereo imaging, and Structure from Motion (SfM).
• Data structure and algorithms, software engineering and design principles, computer systems and microcontrollers.
• Control systems, signals and systems, circuit analysis.
• Calculus I-IV, applied linear algebra, probability, optimization, ODE and PDE, complex analysis.
• Advanced classical mechanics, electromagnetism, quantum mechanics, special relativity.

Undergraduate Machine Learning Research Assistant

Robotics and Control Lab, UBC

Conducted research under Dr. Purang Abolmaesumi on machine learning and computer vision with applications to medical imaging

• Developed and fine-tuned echocardiogram image segmentation pipeline using Florence 2 and SAM 2.
• Benchmarked zero-shot and fine-tuned segmentation performance using MedSAM and SAM 2 on RCL's internal dataset, Stanford's EchoNet-Dynamic, and CAMUS dataset.
• Incorporating Kalman Filter into semantic representation to assist video object segmentation and tracking.

Computer Vision and Robotics Research Scholar

Computer Vision for Smart Structure Lab, UWaterloo

Conducted research under Dr. Chul Min Yeum on Gaussian Splatting neural volume rendering

• Applied spatial and semantic deep learning (SAM) computer vision methods for structural health monitoring.
• Successfully accelerated training speed and imporved rendeirng quality, reflected by a reduction in peak signal-to-noise ratio (PSNR), by incorporating monocular depth estimation and salient object detection to eliminate background.
• Incorporated LiDAR SLAM into Gaussian Splatting to enhance camera pose estimation and aid UAV autonomy mission.

Hardware Engineering Intern

Microchip Technology

Developed expertise in high-frequency analog and mixed-signal IC design and testing

• Conducted successful characterization of a real-time clock (RTC), gaining experience in the design, testing, and analysis of analog and mixed-signal ICs.
• Performed phase-locked loop (PLL) analysis, gaining expertise in frequency and phase synchronization of circuits.
• Conducted return loss analysis, developing skills in signal integrity analysis and design optimization of high-frequency circuits.

Electronics R&D Research Assistant

TRIUMF - Canada's National Particle Accelerator Center

Conducted research under Dr. Thomas Lindner on photomultiplier tubes for neutrino detection

• Successfully developed software enabling electronics mainboard read‐out and data analysis for multi‐photomultiplier tubes (mPMT).
• Through rigorous implementation and testing, our setup was able to capture the emission of a single photon while filtering out the unwanted light source from outside environment.
• Created automated acquisition scripts to enable precise extraction of electronics pulse rates for both calibration and stringent quality control.
• Photomultipliers will be used for world's largest neutrino oscillation detection experiment—Hyper-Kamiokande, in Japan.

Control System Lead - UBC Orbit Satellite Design Team

Currently working in Attitude and Orbit Control System (AOCS) subteam which encompasses attitude and orbit sensors, attitude estimation and guidance, attitude and orbit control algorithms, attitude control actuators for our 1U CubeSat

• Co-developing a Python simulator for satellite attitude control in low Earth orbit, including magnetic actuator model, low-pass filter for onboard sensors, sensor fusion, and satellite system integration.
• Developed optimized firmware for embedded systems using the TMS570 microcontroller, leveraging high-performance DSP library for signal analysis, filtering, and transformation, as well as advanced algorithms such as the Extended Kalman Filter and Sun Model.
• Created training code for DSP library functions in C, providing examples and exercise for new team members.
• CubeSat is scheduled to be launched by the European Space Agency in 2027, who provides training and testing facilities for our team.

Robotics Software Engineer - UBC Rover Design Team

Currently working in the Software subteam which encompasses the rover's autonomous navigation, communication, robotic arm control, and simulation

• Successfully implemented visual-odometry SLAM in ROS 2 using Intel RealSense D415 RGB-D camera for autonomous rover navigation and obstacle avoidance.
• Experimenting with sensor fusion to achieve exceptional localization and mapping performance, including utilizing multiple RGB-D cameras and 2D/3D LiDAR.
• Developing simulations in Gazebo Ignition for rapid navigation software prototyping and Isaac Sim for applying reinforcement learning to control the rover’s robotic arm for various tasks, such as keyboard pressing and switch flickering.
• Scheduled to compete in University Rover Challenge (URC) and Canadian International Rover Challenge (CIRC) in 2025.

PuckPilot Air Hockey Table

Capstone project sponsored by Engineering Physics Project Lab: an air hockey table that integrates computer vision for tracking and reinforcement learning for gameplay decision

• Developing computer vision algorithms for simultaneous puck and mallet tracking using FLIR Blackfly high-speed camera.
• Software system integration using ROS 2 that performs camera calibration, runs computer vision algorithms, deploys agent nodes, handles raw and processed data logging, interfaces with microcontrollers for motor control and feedback, and serves as the central control authority.
• Sim-to-real reinforcement learning for human vs machine gameplay using TorchRL.
• Scheduled to develop other control methods such as Model Predictive Control (MPC) to compare against RL.

Deep Learning-Based Dynamical System Solver

Rethinking deep learning approach for modeling dynamical systems governed by Partial Differential Equations (PDEs) by incorporating both spatial and temporal feature learning.

• Successfully modeled dynamical systems by adding Mamba state space model to learn the temporal features on top of spatial feature extraction, proposing an alternative approach to traditional numerical PDE solvers.
• Developed two varients of our model to learn spatial features, one uses CNN and the other uses image patching, inspired by the patching method in vision transformer.
• Compared to SOTA NeuralPDE, our model achieves competitive mean squared error at half the training time.
• Tested our models using the advection equation and Burgers' equation which both model fluid dynamics, with the latter being a nonlinear second-order PDE.

Autonomous Vehicle Competition

ENPH 353 Engineering Physics Project I: self-driving vehicle in simulated environment using classical computer vision techniques, CNN, and imitation learning.

• Engineered a self-driving vehicle with ROS and performed simulation in Gazebo Classic, incorporating advanced computer vision and machine learning for dynamic text recognition and response to road conditions.
• Implemented imitation learning and real-time camera-based vision systems in the vehicle, enabling sophisticated autonomous navigation and decision-making in complex simulation environments.
• The vehicle was also able to automatically identity various signs next to the road, recognize and interpret the words on the signs for additional points.
• Using a monocular camera as input, our system is able to handle various road condition by dynamically switching between different neural networks we trained and even switching to PID mode in specific conditions.

Autonomous Robot Competition

ENPH 253 Introduction to Instrument Design: self-driving vehicle in real-world setting using inertial odometry, tape sensing, and IR sensing.

• Designed, manufactured, and integrated the mechanical, electrical, and software components of a reaction wheel system; inspired by spacecraft attitude control system, our reaction wheel employed conservation of angular momentum to ensure robot mid-air balance.
• Implemented complementary filter for sensor fusion algorithm using an IMU, utilizing all 9 DoF from the 3-axis gyroscope, accelerometer, and magnetometer to ensure precise robot orientation determination.
• Developed PID algorithm for lane following, acquring tape readings from IR emitter-sensor pair.
• Leveraging Fast Fourier Transform (FFT), our IR sensing system detected the sigal from the IR beacon, filtered out the background IR noise, and ensured our robot follow a straight line in uneven terrain.

Waterloo Engineering Rising Stars Fellowship

University of Waterloo, Faculty of Engineering

The fellowship recognizes exceptional engineering undergraduate students in Canada, offering them an opportunity to engage in cutting-edge engineering research under the mentorship of a University of Waterloo professor. The award includes a stipend of $5,500, along with full coverage of housing and meals, and a 50% reimbursement for travel expenses. Additionally, recipients will be granted a supplementary $2,000 scholarship to support their future graduate studies.