Online Inertial Parameter Estimation for Robotic Loaders

Abstract

Payload estimation is essential to measure productivity, evaluate efficiency in industrial operations and adapting control laws according to the carried weight. One particular problem is to identify how much mass is carried in a mining machine while it is being operated without using strain gauge sensors which require frequent calibration and are prone to failure due to mechanical stress. This paper presents an on-line method to estimate a loader’s payload mass, rotational inertia and viscous friction coefficients employing inertial, torque and speed measurements. The proposed approach introduces a mutual information criterion to select those acceleration and velocity measurements that jointly with the excitation force ensure the identifiability of the parameters. The approach relies on the recursive least-squares algorithm for fast update of the parameters. The proposed strategy is also compared to the implementations based on variants of the least-squares estimator, such as the feasible generalized least squares and the total least squares approach. The approach is tested in simulation and validated in experiments with an industrial semi-autonomous skid-steer loader Cat$^circledR$262C for different loads. Results show that using the recursive least squares it is possible to estimate the parameters with the same level of accuracy than OLS approach, while not needing a large buffer for estimation. Mass is effectively estimated with an RMS error below 1% the total mass of the machine.