PhD Applicant · Robust Perception for Autonomous Systems · Computer Vision

📍 Dhaka, Bangladesh · ✉️ aaaraafaat@outlook.com · 🔗 LinkedIn

Seventeen years flying and instructing in safety-critical aviation — including decision-making in degraded conditions — now aimed at one question: how do we build machine perception that stays reliable when the input does not?

I'm a former fighter pilot and flight instructor (~1300 hrs) completing an MSc in Data Science & Machine Learning (CGPA 4.00 / 4.00). My research is on robust perception in degraded visual environments — estimating when a vision system can be trusted, and characterising why it fails.

A training-free, physics-based approach to predicting per-image object-detection difficulty in adverse conditions such as fog — with no image enhancement or restoration step.

• Scores degradation directly from image physics — dark-channel prior, saturation, contrast, entropy, among other cues
• Built a reproducible feature-extraction and data-cleaning pipeline on the RTTS real-fog dataset (4,322 images)
• Characterises where each cue succeeds and fails — including condition-dependent failure modes where cues are systematically misled by night and coloured-light scenes
• Designed to extend to further degradations (e.g. rain) and additional sensing modalities

Stack: Python · PyTorch · Ultralytics / YOLO · OpenCV · Weights & Biases
Supervisor: Prof. Pintu Chandra Shil, Head, Dept. of CSE, State University of Bangladesh

➡️ Repository → (code, methodology & figures — being prepared for public release)

Flight-training simulator (PT-6 / PC-based AVTD), 2024–2025 — project lead. Integrated COTS hardware and flight-simulation software (Prepar3D, MSFS) with a custom fixed-wing model to deliver affordable, evidence-based ab-initio and part-task pilot training.

VR flight simulator for combat training, 2018 — initiated and advised a virtual-reality part-task trainer; presented the concept to the Chief of Air Staff.

Flight Planning Tool, 2019 — functional manager. Google Maps API with layered custom charts to automate mission flight-chart preparation.

• Robust machine perception in degraded visual environments — object detection and perception-difficulty estimation under fog, rain, and low visibility for autonomous driving and aerial robotics
• Full-stack autonomy — extending perception toward state estimation, control, and planning for autonomous vehicles and unmanned aerial systems
• Simulation, synthetic data, and sim-to-real transfer for safety-critical and human-in-the-loop systems

Python · PyTorch · OpenCV · NumPy / pandas · scikit-learn / SciPy · Ultralytics / YOLO · Weights & Biases · Git · LaTeX / Overleaf

• ✉️ Email: aaaraafaat@outlook.com
• 🔗 LinkedIn: linkedin.com/in/aaaraafaat
• 📄 CV: available on request

_{Best in Flying Trophy, Bangladesh Air Force Academy · United Nations Medal (UNAMID) · TOEFL iBT 103/120 (C1)}