Accurate Real-time Estimation of the Inertia Tensor of Package Delivery Quadrotors

Abstract

The need for quadrotors to provide grasping and payload carrying abilities is ever-growing in several industries. Additional payloads attached to a quadrotor alter its dynamics and eventually affect its control system’s performance. In this work, an accurate real-time estimation of the varying mass and inertia tensor elements of a quadrotor carrying a variable payload is proposed. Parameter estimation is performed via a recursive least squares algorithm that is implemented on the quadrotor’s dynamic model using proper input-output data. Covariance resetting is integrated into the algorithm to increase the convergence rate and accuracy of the obtained estimates. The vertical motion is used to estimate the mass of the system, whereas the rotational motions around the x−, y− , and z − axes are used to identify the elements of the 3x3 inertia tensor matrix. The experiment is designed such that a persistently exciting input is generated to guarantee the convergence of the parameter estimates towards their true values. The proposed identification scheme is validated in numerical simulations and experimentation on a physical quadrotor, the Quanser QBall-2. The obtained results demonstrate the accuracy and convergence rate of the designed estimator, paving the way in front of its integration into an adaptive control system.