Calibration of Stewart platforms using neural networks

Abstract

This paper proposes a novel technique for pose error calibration of Stewart platforms. Traditional calibration techniques use parametric models of the platform, which typically involve either forward or inverse kinematics. The proposed approach divides the entire workspace of a robot into small subspaces. A neural network is utilized to model the pose error within each subspace. There are two major differences between the proposed method and traditional approaches. First, it does not use a parametric model of the platform for error compensation. Instead, it uses a neural network model to approximate the behavior of pose error. The neural network model is then used for error compensation. Second, it does not seek to obtain one set of parameters that works for the entire workspace of the robot. Rather, each subspace of the platform's workspace uses its own set of parameters. The proposed method simplifies the calibration process, and improves pose accuracy. More importantly, the proposed method has the ability to learn and adapt. As a Stewart platform may work in noisy environment and experience shift in its parameters, the adaptive nature of the proposed method makes it more attractive in many applications.