MSc Electronic Engineering candidate, Durham University.
My interests are in sensing and control, including autonomous systems and wearable measurements. 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.
REHAB wearable signal audit examines 4,257 paired records from a public rehabilitation dataset. It compares window, record and exact-content holdouts, then isolates limb and glove signal contributions with a fixed baseline. Published integrity checks and aggregate results are reproducible; a local replay also produces split assignments and held-out predictions.
| Project | What to look at |
|---|---|
| UAV multisensor diagnostics | 35 matched-noise runs isolate timing and trajectory faults, including a local position error missed by the current aggregate gates. |
| Flight-video quality audit | Two released outdoor clips, a 224-frame audit and a 10 Hz follow-up examining a flagged interval and its scene transition. |
| Mission interface | Constrained-language parsing, strict JSON types and regression-tested rejection of ambiguous safety instructions. |
| Thermal validation toolkit | Paired common-mode/reference controls demonstrate which calibration errors cross-channel agreement cannot detect. |
| Project | What to look at |
|---|---|
| Runtime evidence assurance | 360 synthetic telemetry replays measure short-event recall, trigger delay, noisy action episodes and recovery under different thresholds and persistence. |
| Safe neural control certificates | Scalar robust projection with classical baselines; a 126-rollout contraction sweep connects analytic feasibility margins to intervention and cost. |
| Decentralized learning stress test | An eight-peer Adult benchmark with sender-level trimming diagnostics, including tied coordinates and clean/poisoned comparisons. |
| Wireless deadline lab | C++ packet/slot simulation with 160 paired-seed runs comparing clock estimators and uncertainty margins alongside the original four-scheduler benchmark. |
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.