PhD Student · Department of Electrical and Electronic Engineering
I am currently a PhD student in the Department of Electrical and Electronic Engineering, CSP Group, at Imperial College London, supervised by Professor Pier Luigi Dragotti. My research interests lie at the intersection of generative diffusion models, implicit neural representations, synthetic simulation, 3D reconstruction, and immersive VR/AR systems. My current work focuses on developing diffusion models in continuous space.
Prior to Imperial, I obtained my BS & MS in Electrical Engineering at the University of Washington, Seattle, and conducted computer-vision research at the Augmented Cognition Lab NU Boston. Other than academic works, I have also worked in the automotive industry and venture capital, focusing on AI & immersion.
Granted patents and startup projects with deployed customers.
US & CN Patents VR streaming · WebRTC/RTMP/RTSP · 360° Camera · Robotics
A VR teleoperation system that drives a real vehicle from inside a reconstructed virtual environment using a 360° camera, physical wheel/pedal rig, and Wi-Fi/cellular link. Integrates WebRTC-over-RTMP to minimise streaming latency and a robotic-arm rig to synchronise the operator's body movement with the vehicle.
Industry DLP · Adaptive Driving Beam · 1.3M-pixel projection
An automotive lamp that projects AR street signage onto the road and implements a high-precision Adaptive Driving Beam — replacing the conventional 84-pixel matrix LED with a 1.3M-pixel DLP solution compatible with existing automotive platforms. Deployed with customers including Segway, Suzuki, SAIC Motor, and BAIC Motor.
Peer Reviewed Papers
BMVC 2023 British Machine Vision Conference
We introduce TeFS, a method that overcomes the single-viewport limitation of commercial driving simulators (e.g., GTA V) to produce high-fidelity stereo data, and release GTAV-TeFS — the first large-scale GTA V stereo-driving dataset of 88,000+ image pairs with depth, GPS, camera poses, and temporal annotations — used to benchmark state-of-the-art stereo vSLAM models.
ICASSP 2023 IEEE Int. Conf. on Acoustics, Speech and Signal Processing
SAVeS — Scoping Autonomous Vehicle Simulation — is a benchmarking platform that compares simulated environments against real-world counterparts using a unified suite of SLAM-based and learning-based perception models, helping researchers quantify the sim-to-real gap for autonomous ground vehicles.
ACM JATS 2024 Journal on Autonomous Transportation Systems · Vol. 1, Iss. 2, pp. 1–15
Journal extension of our SAVeS platform. Introduces SAVeS+, which uses domain-adaptation to close the gap between synthetic (GTA V PreSIL, CARLA) and real-world (KITTI, nuScenes) data, supporting both geometric SLAM (A-LOAM, LeGO-LOAM, ORB-SLAM3) and learning-based perception models. Models trained on processed synthetic data via SAVeS+ match real-data baselines at equal scale.
ICRA 2025 IEEE Int. Conf. on Robotics and Automation · pp. 8412–8419
We introduce DcTDM, an open-source dual-conditioned temporal diffusion model for generating realistic, temporally consistent long driving videos. Built with DriveSceneDDM, a dataset with scene text, depth maps, and Canny edges, DcTDM produces driving videos up to 40 seconds and improves consistency and frame quality by over 25%.
3DV 2026 Int. Conf. on 3D Vision · arXiv:2512.14406
State-of-the-art dynamic novel-view synthesis methods often fail under significant viewpoint deviations. We introduce Expanded Dynamic NeRF (ExpanDyNeRF), a monocular framework that combines Gaussian-splatting priors with a pseudo-ground-truth generation strategy to enable realistic synthesis under large-angle rotations. We also release SynDM — the first synthetic multiview dataset for dynamic scenes with explicit side-view supervision, captured via a custom GTA V multi-camera pipeline.
Selected hands-on engineering, mobile, robotics, and game-design projects from my earlier work.
VR · Unity · Multi-user Simulation
Multi-user VR cycling system supporting up to 20 simultaneous players, combining custom steering hardware, ultrasonic speed sensing, and a Unity open-world town.
UW News Spotlight · Robotics · Computer Vision
3D-printed robotic hand that plays piano from scanned sheet music using Optical Music Recognition, and accepts remote control through flex-sensor gloves over BLE.
Augmented Reality · Mobile
iOS app that detects up to five paintings in real time and replaces them with custom imagery — no prior training required.
Embedded · EEG · BLE
Five-mode robotic arm controlled by flex-sensor glove and EEG concentration, built on Arduino Uno + Mega with Bluetooth LE communication.
Unity · WebGL · Mobile
Fast-paced arcade maze runner with 30 hand-designed levels, time-pressured irreversible movement, and pickups (springs, invincible pills).
Gameduino 2 · Sprite Game · Embedded
Arcade flight shooter for the Gameduino 2 platform, featured by the Gameduino creator. Touch controls, escalating enemy patterns, bombs and lasers.