FPGA implementation of current-based joint torque estimation in antagonistic tendon-driven actuators

Abstract

Antagonistic tendon-driven actuators are found in various robotic systems due to their remote-actuation capability and inherent backlash compensation. These actuators typically employ non-backdrivable drives and rely on external force or torque measurements for compliant control. Using proprioceptive, force-transparent drives could enable joint-torque estimation based on motor current, allowing fault detection and sensorless control. Consequently, this thesis analyzes current-based tendon-force and joint-torque estimation on a modular testbed and a FPGA-based control platform. Suitable friction models required for estimating tendon-tension were identified and incorporated into a reaction-force observer structure. Furthermore, polynomial approximations suitable for FPGA implementation were developed for the kinematic relations required to estimate joint position and joint torque. Based on the results, a drive- and joint-control scheme was proposed, that is capable of realizing sensorless impedance control on FPGAs.