A New State Estimation Framework for Humanoids based on a Moving Horizon Estimator

Abstract

A large number of humanoid robot state estimators based on the Kalman filter (KF) have been proposed. However, such estimators cannot guarantee optimality when the system model is nonlinear or when non-Gaussian/correlated modelling error is present. Additionally, the use of a KF causes difficulty in coping with equality or inequality constraints. Because a bipedal humanoid robot is a complex nonlinear system, its mathematical model is limited in its ability to express the system accurately. Therefore, KFbased humanoid state estimation has unavoidable limitations. To overcome these, we propose a new approach to humanoid state estimation using a moving horizon estimation theory. So far, there are almost no studies on the moving horizon estimator based humanoid state estimator, which is capable of accommodating nonlinear systems and constraints as well as being more robust to the nonGaussian/correlated modelling error. The proposed estimator framework facilitates the use of a simple model, even in the presence of large modelling error. Additionally, it can estimate the humanoid state more accurately than the existing KF-based estimator framework. The performance and characteristics of the proposed approach were confirmed experimentally.