Working with Dr. Steve McGuire at UCSC's HARE Lab and Dr. Leilani Gilpin at the AIEA Lab, our goal is to develop a robust, unified MTRL controller capable of smoothly switching between multiple behaviors using Nvidia's IsaacSim and IsaacLab.

Using ROS2 and Gazebo, we aim to validate our controller on a physical Unitree B1 quadruped and demonstrate real-world robustness through sim-to-real transfer.

Implemented a geometric controller and polynomial trajectory optimizer for a drone racing competition as part of ECE 264: Visual Navigation for Autonomous Vehicles (based on MIT 16.485).

We finished the obstacle course in 8.5 seconds — 1.8 seconds faster than MIT's record of 10.3s as of March 2026

Built with SvelteKit, all the contents and animations that you see are hand-crafted by yours truly with SCSS or Procreate :)

No AI was used in any part of this website.

A WYSIWYG graphical neural network maker. It lets you to create and train simple neural networks on custom datasets without any code, which can then be exported to be deployed elsewhere.

Using a self-built templating engine, DeepFusion uses no UI frameworks and has minimal performance overhead. It's also won first place in over 7 competitions, including the Congressional App Challenge, Conrad Challenge, and Raspberry Pi Coolest Project competition.