External torque virtual sensors applied to a 6-DOF robot arm

Robotic manipulators are arguably the most important engineering feats for future automation, enabling in-space construction, automating maintenance tasks, and advancing industrial manufacturing. Further advancements in simulation software, computational power, machine learning approaches, and emerging trends of cyber-physical systems are increasing the demands for data to improve control and safety systems onboard. A supplement to these demands is virtual sensors or soft sensors, which produce signals like physical sensors based on a predefined architecture and algorithm. Virtual sensors encompass topics such as sensor fusion algorithms and estimation theory but also enable measurement of properties that lack a physical sensor counterpart. Some advantages of virtual sensors that can benefit robotics include redundancy, reliability, adaptability, and cost reduction. All of these properties stem from the fact that there is no physical hardware onboard the robot, ultimately avoiding wear, drift, and maintenance. This paper leverages virtual sensors as an external torque sensor at the end-effector using a combination of the system’s dynamics and external observers, specifically the Unscented Kalman Filter (UKF), the first order-momentum observer (FOMO), and the General Momentum Kalman Filter (GMKF), on a simulated six-degree-of-freedom UR5 robot arm.