Selected Projects

Research & Engineering

Projects spanning deep learning, computer vision, reinforcement learning, and robotics. The first three have dedicated detail pages with full methodology, figures, and results.

01

Five deep learning architectures evaluated on a custom synthetic dataset (MiniMarket77) of 12,000 cluttered scenes for real-time robotic pick-and-place. Best model (PointWeb) achieved 0.986 mIoU; compression study reduced latency to 1.77 s with INT8 quantisation.

PyTorchPointNetDGCNNPointWebStratified TransformerCUDA
02

Prompt-Based Image Segmentation with SAM2

MSc Project · Oxford-IIIT Pet Dataset · Jan – Mar 2025

Interactive semantic segmentation application using SAM2 with point and bounding box prompts. Achieved 89% pixel accuracy and 0.917 Dice coefficient on Oxford-IIIT Pet, a 63% IoU improvement over the baseline UNet.

PyTorchSAM2UNetGradioOpenCV
03

Smooth, precise pneumatic actuation system for delicate robotic gripping using a cascade PID + Unknown Input Observer controller. Full mathematical modelling, multi-domain simulation, physical fabrication, and experimental validation.

MATLABSimulinkSimscapeBeagleBone BlackPIDObserver Design
04

End-to-End RL for Autonomous Racetrack Navigation

MSc Project · University of Edinburgh · Jan – Mar 2025

Progressive RL pipeline from Value/Policy Iteration to DDPG with Prioritised Experience Replay, adaptive noise, and orthogonal initialisation. Best configuration from a 60+ model sweep achieved a mean return of 374.86 on Racetrack-v0.

PyTorchDDPGDQNGymnasiumhighway-envPER
05

Emotional Expression Development in Infants via Deep Learning

MSc Project · ResNet-18 · Jan – Mar 2025

Cross-age dataset combining Tromsø Infant Database with AffectNet (26,000+ images across 5 developmental stages). Two ResNet-18 models demonstrated a nonlinear developmental trajectory — adult-like expressions emerge in late childhood.

PyTorchResNet-18ResNet-50SMOTEOpenCV