Hierarchical mixture of experts for autonomous unmanned aerial vehicles utilizing thrust models and acoustics

Abstract

Accurate position, velocity, attitude, and angular velocity state estimation is crucial for unmanned aerial vehicles, especially in enabling them with autonomous capabilities. It is necessary to adequately model and account for all the environmental and dynamic flight parameters. A hierarchical mixture of experts (HME) framework has been viable in improving state estimation accuracy in interplanetary orbit determination problems, and this paper proposes an extension for quadcopters. It is shown that the state and motor angular velocity estimation accuracy can be significantly improved by processing different thrust models, and acoustic parameters have an important, previously unreported, role in this improvement. Higher motor angular velocities produce higher noise levels, and thus, the relationships of the onboard acoustic measurements to the vehicle state parameters play an essential part in estimation. The motor angular velocities’ estimations depend on the extended Kalman filter solutions or an acoustic curve fit. The experts in the HME framework utilize the state estimation solutions from the extended Kalman filters and the motor angular velocity estimations to compare against the telemetry data as truth. The overall HME solution is compared against a non-acoustic static thrust model. Illustrative examples and analysis presented in this paper reveal that the proposed estimation solutions can also apply to other flight vehicles for onboard real-time implementation to leverage autonomy.