MSc Electronic Engineering candidate, Durham University.
My interests are in sensing and control for autonomous systems. I focus on what changes when the inputs are imperfect: timestamps drift, measurements become unreliable, or a learned controller encounters model uncertainty.
The research portfolio brings together UAV field footage, software experiments and analysis. The individual projects are below.
| Project | What to look at |
|---|---|
| UAV multisensor diagnostics | A fault-injection case with nearly normal average sensor rates but 18.314 ms p95 camera–LiDAR timing mismatch. The analysis checks timing, sequence loss and trajectory error separately. |
| Flight-video quality audit | Analysis of two released outdoor clips: 224 sampled frames, with two low-sharpness samples in Clip B. Includes the videos, frame metrics and timeline. |
| Mission interface | A rule-based parser and structured checks for speed, clearance and fallback behavior. The contract makes the decision available for inspection before execution. |
| Thermal validation toolkit | Synthetic multichannel data with offset, drift and status faults; separate channel and reference metrics help explain each failure. |
| Project | What to look at |
|---|---|
| Runtime evidence assurance | Synthetic telemetry replay connecting monitor persistence, stopping margins and fallback recommendations. |
| Safe neural control certificates | A learned policy with an analytic robust projection for a scalar sampled-data model; classical baselines and counterexamples show where each guarantee comes from. |
| Decentralized learning stress test | An eight-peer Adult benchmark examining the tradeoff between clean-data accuracy and resistance to poisoned updates. |
| Wireless deadline lab | A C++ packet/slot simulator comparing four schedulers. Clock-aware EDF helps in bursty-channel cases but loses to FIFO in the configured clock-holdover case. |
Each project links its results to the data, configuration and commands used to produce them. My current research question is how timing and model uncertainty should change a system's decision to proceed, constrain an action or request better evidence.