Improving SEA Joint Torque Sensing for Enhanced Torque Estimation in Human-Machine Interaction

Estimating human-robot interaction (HRI) in rehabilitative robotics poses challenges from both mechatronics and control standpoints. In this study, the authors introduce a method to tackle non-linear effects impacting Series Elastic Actuator (SEA) mechatronic systems. Achieving an accurate characterization of the SEA system enables improved torque estimation based on spring deformation. By improving torque sensing capabilities, the objective is to enhance HRI assessment for patient evaluation and enable compliant closed-loop control. The adopted method involves experimental testing, subjecting the SEA to mechanical stress under various load conditions. Experiments were conducted to explore SEA characteristics across different load configurations: joint at the bench with no-load, low-medium-high load, and joint assembled on an exoskeleton prototype. Analysis of the data revealed variability in SEA torque estimation errors, with higher loads associated with greater errors. To mitigate these errors and enhance torque estimation, the non-linearities have been characterized using two Fourier series (5th and 8th order) in both no-load and load conditions. Theoretical approaches and experimental results of this SEA characterization methodology will be presented to demonstrate the feasibility of this approach.