Certified Randomised Algorithms for Multi-channel Airborne Radar

Standard computer vision can segment and label objects in a scene, but safety-critical defence and security applications require a deeper, physically grounded understanding. This PhD project addresses a fundamental question: can an autonomous sensor infer governing physical laws directly from raw observations? 

Focusing on high-impact applications, such as tracking and characterising autonomous drones, this research will establish an end-to-end framework mapping raw video to segmented objects and, crucially, to their underlying differential equations. By identifying the dynamical laws that dictate how an object moves or deforms, the system will infer intrinsic physical parameters (such as mass and dimensions), classifying targets based on their physical behaviour rather than appearance alone.  

Moving beyond equation discovery, the project investigates operational robustness against deceptive maneuvers. It will evaluate how the pipeline responds to adversarial physical behaviors and develop multi-modal sensor fusion, integrating complementary modalities such as radar, to mitigate these vulnerabilities. 

This project will leverage physics-informed machine learning, dynamical systems, and robust multi-sensor perception to deliver explainable, tamper-resistant AI for modern defence challenges.