Harnessing multi-modal deep learning for multi-drone navigation-based trajectory prediction system

In recent times, unmanned aerial vehicles (UAVs), also known as drones find useful in different application areas like military, healthcare, logistics, image and video mapping, precision farming, wireless communication and aerial surveillance. Research works on multi-drone collaborative flight path planning has drawn extensive attention in the domain of drones, representing singular benefits in large-scale monitoring, difficult task performance, and disaster response. Among the core technologies in multi-UAV collaborative operations, advancements in trajectory planning play a crucial role in ensuring the safety and efficiency of these coordinated missions. A trajectory design of the drone certainly plays a significant part in the application performance, efficiency, and development. By leveraging artificial intelligence (AI) and machine learning (ML), UAVs can effectively perceive their surroundings and make more advanced decisions. This paper proposes a navigation-based Trajectory Prediction System using Multi-Modal Deep Architecture (NTP-MMDA) model. The primary intention of the NTP-MMDA model is to develop an intelligent multi-drone navigation system for accurate trajectory prediction to ensure coordinated path planning and collision avoidance. At first, the NTP-MMDA technique performs data pre-processing by using the quantile normalization method to ensure uniformity. Furthermore, the trajectory prediction process is mainly executed by three models, such as bidirectional gated recurrent unit (BiGRU), variational autoencoder (VAE), and adaptive deep belief network (ADBN). The comparison study of the NTP-MMDA approach portrayed a superior performance, achieving the lowest MSE of 0.0021 among all models under the Drone Trajectory dataset.