Research
Side Projects
Research
Shape n' Swarm: Hands-on, Shape-aware Generative Authoring with Swarm UI and LLMs
Matthew Jeung, Anup Sathya, Michael Qian, Steven Arellano, Luke Jimenez, Ken Nakagaki
UIST 2025 Best Demo Award (Jury's Choice)
Paper | Video | 30s Video
CubeLocomotion
User manipulation-based robotic control using deep reinforcement learning.
Research In Progress (Leading project)
View details ▼
Progress Snapshots
I designed a robust cube-shaped robot that can flexibly locomote using a single embedded actuator, and created created a MuJoCo simulation of the hardware. I have a decent PPO control policy that enables basic flipping, turning, and translational movement.
Early hardware testing for cube rotation via vibration.
Policy learns to flip from horizontally-oriented flywheel to vertically-oriented flywheel.
Skills Involved
Robotics Simulation, Deep Reinforcement Learning, Hardware Design (PCB design, 3D modelling)
Relevant Research Domains
Deep Reinforcement Learning, Self-Reconfiguring, Modular Robotics
Initial sketch of idea.
Motivation
This project pushes towards a new form of shared autonomy, where user directly defines the form and functionality of the system. It was inspired by past work on self-reconfiguring, modular robotic systems, such as M-Blocks. My goal for this project is to allow users to assemble shapes with cube primitives, which immediately know how to move based on a shape-flexible control policy.
Haptics Project
Modular toolkit for assembling haptic devices.
Collaboration with Sony, submitted to CHI 2026 (Co-first author)
Paper | PW for reviewers: review
Role
Working as a co-lead author on a haptics paper. Very excited about the project, but hidden for now!
Relevant Research Domains
Haptics
Skills Involved
3D Modelling, Electronics (microcontrollers), Programming, Simulation
Unsupervised Action Representation Learning
Developing an autoencoder / world model framework.
Research In Progress (Working under a PhD)
View details ▼
Role
Working under a PhD student, I have contributed by training and deploying RL agents, visualizing model training and outputs, working on model architecture, and completing other tasks like dataloading.
Relevant Research Domains
Representation Learning, World Models, Diffusion, Robotics
Skills Involved
PyTorch, CUDA
Side Projects
GreenGuide
Grocery store item sustainability feedback through barcode-scanning glove.
MakeHarvard 2025 Most Sustainable Project
Skills Involved
Electronics (Raspberry Pi, Soldering, etc), 3D Modelling (Fusion 360), Software Development (Python, Flask, APIs)
Description
GreenGuide provides instant sustainability feedback for grocery items using a barcode-scanning glove, OpenFoodFacts API, and LLM agent.
1lb Combat Bot
Horizontal spinner bot designed completely from scratch. Features weapon belt system and asymmetrical blade.
UChicago Robotics | Video
Skills Involved
3D Modelling (Fusion 360), Electronics (Vertiq drone motor, ESCs, drive motors)
Description
A competitive robot built for 1lb combat, featuring a custom weapon belt and unique blade design for optimal performance. Made with one partner.
senseMEE
ML-based Spotify extension to autoqueue songs based on physical activity, environment, and weather.
Github | Video
Skills Involved
Mobile App Development (Swift, Spotify API, IMU data filtering), Machine Learning (scikit-learn)
Description
senseMEE uses machine learning to analyze physical activity, environment, and weather, autoqueuing songs on Spotify to match your context.