Yuan Fang

I'm a master student at Columbia University (Expected 2026). I am working on robotics under the surpervision of Prof. Yunzhu Li. I am also fortunate to be advised by Dr. Hu yuhang and Prof. Hod Lipson. Before entering Columbia, I earned B.S from Shanghai Jiao Tong University.

Email: yf2716 at columbia dot com.

Research

I'm interested in scalable learning methods for robots. For robot hardware, I am especially interested in life-like humanoid robots.

Realistic Gaze Transformer (RGT), a Transformer-VAE framework that learns full-head gaze dynamics, including head rotations and eyelid movements, from human motion capture data.

Projects

Process Scheduler: Designed and implemented a custom kernel scheduler that optimizes task completion in multi-agent environments. It uses weighted process prioritization and enhanced multi-core load balancing, outperforming the default CFS in both average and tail latency. Performance evaluation was conducted using bpftrace in virtualized environments.

Page Table Real-Time Inspection: Built a Linux kernel tool for real-time virtual-to-physical address mapping inspection. Introduced a "shadow page table" mechanism and custom syscalls to enable one process to track another’s VMAs and PTEs. Leveraged eBPF and MMU notifiers for low-overhead memory change monitoring.

Soft-Cloth Simulation: Simulated deformable objects using C++ and mass-spring models.

Stable Fluid Simulation: Developed tools for fluid motion using Navier–Stokes solvers and MAC grid stabilization.

Implemented PPO-based RL policy for robotic hand to pick up lug nuts using vision and tactile feedback.

Developed a Java maze game where players collect keys and navigate through multiple custom-designed levels in pixel-art style.