Currently, a CS student at Simon Fraser University and a robotics research assistant in the MARS (Multi-Agent Robotic Systems) Lab, supervised by Prof. Mo Chen,
where I work on multi-agent robotic safety, Hamilton–Jacobi reachability, and model predictive control.
My work tends to start from theory (reachability, critics, planners) and end with something that actually runs;
whether that’s a JAX-accelerated MPPI controller (multi-agent-mppi) rolling out trajectories on a GPU,
or multiple differential-drive robots executing safety-aware policies.
At this very moment, I’m interested in making formal verification-based safety (i.e. HJ) learnable online, with some robustness and anti-forgetting mechanisms.
so that embodied, reset-limited robots can adapt their safety structure over time without losing rigorous safety and liveness guarantees.
More broadly, I’m also interested in RL-based control and computer vision, since believing that a single method can address all problems is a fallacy ;).
I also dabbled on the biomedical side! I am grateful to have worked with Prof. Ghassan Hamarneh and his former postdoc student, Dr. Ben Cardoen (a super cool guy!),
in SFU’s MAI (Medical Image Analysis) Lab,
building physics-aware 3D SMLM simulators and workflows for protein structure recovery,
as well as tooling for Federated Learning on bimodal data and HPC.
Outside academia,
I’m a heavy Hacker News lurker, browsing interesting topics using my in-house static Hacker Dash [0] app daily.
I’ve also wired up LLM + RAG systems with vector databases (a homebrew local RAG LLM with pgvector, Ray, LangChain, llama_index [*]),
and spent too many evenings shipping small, useful-in-house, yet-another-x-does-y tools and toys:
a personal tmux fork (tmuxx [*]),
a tiny Unix shell in C (kell [*]),
an Org-mode/Markdown link generator (org-link [*]) that parallel-scrapes page titles,
and a terminal-native AI sidekick (chatline [*]) (since there were way too many options out there and I just want to use curl).
On the ML/vision side, along with university courses, I’ve worked on a Sensory Data Classification project using Spark + VAE classifier,
an Any Images as Panorama OpenCV stitcher,
and 3D CV pipelines for 3D Reconstruction and Augmented Reality [*].
And between 2021 and 2022, a trail of some hand-rolled education-only neural networks nn-1 (numpy+CUDA [*]) thanks to Michael Nielsen’s amazing Neural Networks and Deep Learning book,
and ofc, I later JAX-ed-them-all, a vgg16-jax [*] and a trans-jax [*] implementation (one for an OCR app, another for a character-level Transformer project; both for recreational learning).
For fun, I also designed game AIs such as Nine Men’s Morris Perfecto [*] and Tic-Tac-Menace-Toe (MENACE RL) [1],
and even once went down the rabbit hole of CPU Design on an FPGA (Altera DE10-STD) with a 4-bit machine and custom opcodes [*].
I like problems that force me to care about both mathematical structure and systems details — and I’m happiest when there’s a profiler open and some idea I don’t yet fully understand.
[*] (I promise that one day I’ll organise all of my projects in one place. nicely, for real.)
<small>
Give me i3, tmux, and Emacs with org-mode, and I’ll be happy on a desert island forever after ;D
(this entire site’s CI/CD is Emacs in a container \[T]/)
</small>
this site doesn’t collect anything; its very ae-static ;)
Publications & Works
\(\triangleright\) K. K. Yang, G. Lee, N. Jahanshahi, and M. Chen. “Dynamics-Aware Buffered Time-to-Reach Collision Avoidance using Hamilton–Jacobi Reachability.”Preprint, to be submitted to IEEE Robotics and Automation Letters, 2025.
<tiny>NOTE: the link to preprint is coming very soon!!!</tiny>
TO INTERNET, BUILD FROM SCRATCH WITH LOVE AND EMACS \[T]/ [2025-12-24 Wed 00:59]