A Model Inversion Procedure for Control of Nonlinear Series Elastic Actuators

Area of Science:

• Robotics
• Control Systems Engineering
• Materials Science (Viscoelasticity)

Background:

• Accurate modeling of compliant elements is essential for advanced robot control.
• Viscoelastic materials in Series Elastic Actuators (SEAs) present complex, non-linear challenges for analytical inversion.
• Model inversion is critical for enabling precise model-based control in human-robot interaction (HRI).

Purpose of the Study:

• To develop and evaluate an accurate model inversion procedure for viscoelastic compliant elements in rotary SEAs.
• To improve the precision of model-based control for physical HRI applications.
• To address the non-trivial analytical inversion of complex non-linear terms in elastomeric SEA models.

Main Methods:

• Coupling a partially analytical inverse model with a disturbance observer (DOB).
• Implementing and comparing inversion with and without a DOB, using two filter types.
• Quantifying inversion accuracy by comparing desired setpoints to 'actual setpoints' derived from the forward model.

Main Results:

• The proposed inversion procedure using a DOB consistently yielded lower root-mean square (RMS) errors.
• Performance was superior compared to inversions omitting the DOB, across various torque signals.
• The DOB-enhanced method demonstrated greater accuracy in tracking desired torque signals.

Conclusions:

• The coupled analytical inverse model and DOB provide a mathematically tractable and accurate inverse for complex viscoelastic elastomer models.
• This approach enhances the feasibility of precise model-based control in HRI involving SEAs.
• The findings suggest a significant improvement in the fidelity of SEA control systems.