Mass Estimation in Manipulation Tasks of Domestic Service Robots using Fault Reconstruction Techniques

TL;DR

This paper introduces a novel method for estimating the weight of objects during manipulation tasks in domestic service robots using Sliding Mode Observers, enhancing object recognition capabilities beyond visual data.

Contribution

The work presents a new approach to estimate object weight in domestic robots through fault reconstruction techniques, specifically using Sliding Mode Observers, which has not been explored before in this context.

Findings

Successful weight estimation in simulation and real robot tests.

Improved object recognition accuracy with physical property estimation.

Potential for enhanced manipulation performance in domestic robots.

Abstract

Manipulation is a key capability in domestic service robots, as can be seen in the rulebooks of last Robocup@Home editions. Currently, object recognition is performed based mostly on visual information. Some robots use also 3D information such as point clouds or laser scans but, to the knowledge of authors, robots don't use physical properties to improve object recognition. Estimation of an object's weight during a manipulation task is something new in the @Home league and such ability can improve performance of domestic service robots. In this work we propose to estimate the weight of the grasped object using Sliding Mode Observers. If we consider the manipulator without load as the nominal system and object's weight as a fault signal, we can estimate such weight by an appropriate filtering of the output error injection term of the sliding mode observer. To implement our proposal we used MATLAB and Simulink Robotics System Toolbox, ROS Toolbox and Simscape. To improve computation time we exported all algorithms to standalone ROS nodes from Simulink models. Tests were performed using two platforms: Justina's left manipulator (a robot developed at Biorobotics Laboratory, UNAM) and Neuronics Katana manipulators. We present results in simulation and discuss the performance of the proposed system and the possible sources of error. Finally we present our conclusions and state the future work.