Data-Driven and Explainable Artificial Intelligence Modelling for Quadrotor Crash Area Prediction

The advent of quadrotors has revolutionized applications such as surveillance, logistics, and disaster response, owing to their versatility and manoeuvrability. However, their nonlinear dynamics and sensitivity to actuator and propulsion failures pose significant safety risks. Additionally, the blackbox nature of traditional artificial intelligence (AI) models hinders transparency and trustworthiness in safety-critical predictions. This paper presents a novel data-driven and explainable AI framework for predicting quadrotor crash areas under single actuator and complete power failure scenarios. The framework uses high-fidelity simulation data and the Feyn QLattice algorithm to model complex descent dynamics while offering an interpretable symbolic expression for external stakeholders. Comparative predictive evaluations with machine learning models, including random forests (RF) and extreme gradient boosting (XGB), reveal that the QLattice algorithm achieves competitive accuracy with RMSE and R<sup xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">2</sup> values of 7.666 and 0.969, respectively. Validated through 5-fold crossvalidation and hold-out testing, the framework demonstrates its potential to advance quadrotor safety by balancing accuracy, efficiency, and interpretability. Upcoming research will focus on integrating wind disturbances, investigating additional failure scenarios, and creating and refining interpretable algorithms to improve predictive performance.