Training a robotic arm to estimate the weight of a suspended object

Abstract

Empowered by machine learning (ML) and statistics, researchers are able to leverage a larger amount and variety of data to facilitate research processes. Moreover, computational tools can mine and translate insights from data for multiple purposes. Here, we present a new approach to estimating the weight of an object using robotic torque data. Our focus is on processing and analyzing raw torque data collected from the joints of a robotic arm that was originally intended for robotic movement and security. We demonstrate that by applying ML models using scikit-learn, it is possible to overcome the limitations of summary statistics and accurately predict the weight of an object held by the arm. Our work provides a practical example of how ML can be applied to data science problems beyond simple curve fitting, and we provide a workflow guideline to make this approach more accessible to researchers outside the ML field.