Position prediction using disturbance observer for planar pushing

TL;DR

This paper introduces a disturbance observer-based method for predicting the position and orientation of objects pushed on a plane, leveraging linear relationships between forces and disturbances for improved accuracy.

Contribution

It formulates the push prediction as a disturbance observer design problem and develops a two-phase algorithm combining identification and prediction.

Findings

Accurately estimates disturbances affecting the object.

Shows improved prediction accuracy over simple correction methods.

Validates the approach with experimental data from MIT MCube Lab.

Abstract

The position and the orientation of a rigid body object pushed by a robot on a planar surface are extremely difficult to predict. In this paper, the prediction problem is formulated as a disturbance observer design problem. The disturbance observer provides accurate estimation of the total sum of model errors and external disturbances acting on the object. From the estimation results, it is revealed that there is a strong linear relationship between the applied force or torque and the estimated disturbances. The proposed prediction algorithm has two phases: the identification & the prediction. During the identification phase, the linear relationship is identified from the observer output using a recursive least-square algorithm. In the prediction phase, the identified linear relationship is used with a force plan, which would be provided by a mission planner, to predict the position and the orientation of an object. The algorithm is tested for six different push experimental data available from the MIT MCube Lab. The proposed algorithm shows improved performance in reducing the prediction error compared to a simple correction algorithm.