AI-Based Approach to Simulate the Drone Dynamics in Three-Dimensional Space

Abstract

ABSTRACT Unmanned aerial vehicles, often known as drones, have unlocked new possibilities in a range of industries, including surveillance, transportation, construction, and agriculture. Analysis of drone behavior is challenging due to the complex interplay of several factors, including speed, altitude, orientation, and trajectory. Simulations of drone dynamics are an important requirement for many fields because simulations allow researchers to create and test drones in complex scenarios that might be challenging or unsafe. There are numerous proposed drone dynamic models on the base of Newtonian and fluid dynamics. These models include a variety of model parameters like force, gravity, propeller characteristics and air density. It is not feasible to examine the necessary model parameters to replicate a particular drone due to these parameters, but the suggested model can simulate any generic drone. An AI-based approach can be utilized to model drone dynamics simply compared to the Newtonian and fluid dynamics methodologies. It involves an advanced AI model trained on huge datasets spanning multiple real-world flight scenarios. The datasets comprised a broad range of maneuvers, including the eight, circular, and lazy-eight patterns chosen to represent a variety of types of drone motions. Multiple approaches were employed in model creation, including multi-output regression, support vector machine, neural network, and convolutional neural network (CNN). Among these, the CNN model displayed the highest accuracy, achieving 78%. The quantitative validation approach was accomplished by comparing predicted maneuvers versus real-world drone maneuvers. Future work involves further development of a CNN-based drone dynamic model combined with a virtual reality environment.