Numerical Optimisation-Based Control Pipeline for Robot Arm for Machine Learning Experiments

Abstract

In this paper, a control pipeline for an industrial robot arm is presented. The robot is proposed as a test bench for machine learning experiments, where it is important to have a reliable and repeatable behavior of the robot while its control system needs to be flexible. The proposed controller is model-based, and is implemented as a quadratic program, imitating feedback linearization and computed torque controller with stable control error dynamics. The paper discusses practical aspects of the software suit developed for the project and presents a favorable set of the third party packages which allowed to create a fast and reliable implementation of the proposed control scheme. The properties of the proposed control were studied using simulation in the Gazebo environment.