Data Driven Modeling and Control of Delivery Drone

Data driven dynamic modeling and control of a quadcopter with slung load (i.e., Delivery drone) is developed in this study. The dynamic and controls are studied with both the physics-based mathematical model and data driven modeling methods for comparison purpose. The delivery drone is modeled as 2D nonlinear quadcopter with slung load, based on Lagrangian formulation, from a previous study. Initially, an LQR control is developed with the linearized mathematical model to analyze the performance of delivery drone. For a small disturbance, the oscillations of degrees of freedom associated with quadcopter take a shorter time to reach the equilibrium with LQR control. However, the oscillation of slung load takes longer time to decay than quadcopter oscillations. In the next stage, the data driven model based on feedforward neural network is developed for nonlinear dynamic model of delivery drone. The input data is generated with the numerical simulations of nonlinear model of delivery drone. The simulated response data of slung load oscillation, calculated for small disturbances, is found to be unbounded and unstable. This shows that stability analysis of delivery drone play a crucial role in generating the stable input data. In the next stage, a PID controller is designed with the Data Driven model of the delivery drone. The parameters of the PD controller are found using the trial and error approach. The initial results show the PD controlled system is efficient in controlling the system dynamic oscillations for initial disturbances. These initial results show that the controller design based on Data Driven model can be effectively used in the delivery drones without the necessity of physics-based mathematical models. However, further studies are needed in the data driven-based dynamic modeling and control design with advanced neural network approaches, and experimental validation of results.