A Multi-task Transformer Architecture for Drone State Identification and Trajectory Prediction

With the proliferation of unmanned aerial vehicles in various applications ranging from delivery services to surveillance and rescue operations, an efficient drone state identification and trajectory prediction framework is becoming a mandate. This work introduces a novel multi-task learning framework that provides drone state identification and trajectory prediction enhancements with the use of a novel Transformer neural network architecture. The proposed system exploits the ability of Transformer models to handle sequential data, capturing more effectively than conventional models' temporal dependencies and relations inherent to drone movement data. The proposed framework utilizes a dual-task learning approach, facilitating simultaneous prediction of drone state and trajectory and enhancing both tasks' performance via shared feature learning. Extensive evaluations are conducted to validate the efficiency of the proposed framework, with two complementary datasets encompassing diverse drone movements and conditions. The results demonstrate significant improvement in terms of drone state identification and trajectory prediction performance compared to existing methods.