[Paper Review] Dynamic Control of Pneumatic Muscle Actuators

Isuru S. Godage, Yue Chen+1 more|arXiv (Cornell University)|Nov 12, 2018

Soft Robotics and Applications9 references8 citations

TL;DR

This paper proposes a dynamic control framework for pneumatic muscle actuators (PMAs) using a Bouc-Wen hysteresis model to improve position tracking performance under external load. By integrating the hysteresis model into a computed torque controller, the method achieves significantly better phase response and tracking accuracy than standard PID control, especially at frequencies up to 2 Hz, demonstrating strong potential for high-bandwidth soft robotics applications.

ABSTRACT

Pneumatic muscle actuators (PMA) are easy-to-fabricate, lightweight, compliant, and have high power-to-weight ratio, thus making them the ideal actuation choice for many soft and continuum robots. But so far, limited work has been carried out in dynamic control of PMAs. One reason is that PMAs are highly hysteretic. Coupled with their high compliance and response lag, PMAs are challenging to control, particularly when subjected to external loads. The hysteresis models proposed to-date rely on many physical and mechanical parameters that are difficult to measure reliably and therefore of limited use for implementing dynamic control. In this work, we employ a Bouc-Wen hysteresis modeling approach to account for the hysteresis of PMAs and use the model for implementing dynamic control. The controller is then compared to PID feedback control for a number of dynamic position tracking tests. The dynamic control based on the Bouc-Wen hysteresis model shows significantly better tracking performance. This work lays the foundation towards implementing dynamic control for PMA-powered high degrees of freedom soft and continuum robots.

Motivation & Objective

• Address the challenge of controlling highly hysteretic, compliant PMAs in dynamic, high-bandwidth applications.
• Overcome limitations of traditional PID control due to inherent phase lag and nonlinearities in PMAs.
• Develop a practical, experimentally identified dynamic model that captures PMA hysteresis without relying on hard-to-measure physical parameters.
• Enable dynamic, closed-loop control of PMAs under external load for use in multisection continuum robots.
• Demonstrate the feasibility of using computed torque control with a hysteresis model for real-time, high-performance tracking.

Proposed method

• Employ a Bouc-Wen hysteresis model to capture the nonlinear, rate-dependent behavior of PMAs.
• Identify model parameters through experimental characterization of PMA response under varying pressure and external load.
• Integrate the hysteresis model into a computed torque control framework for feedforward compensation of dynamic nonlinearities.
• Implement the controller in a Simulink Realtime environment with a 100 Hz inner loop and 20 Hz outer loop for stable pressure regulation.
• Use a high-resolution linear optical encoder for real-time position feedback in closed-loop control.
• Compare the computed torque controller against a standard joint-space PID controller under identical experimental conditions.

Experimental results

Research questions

• RQ1 Can a Bouc-Wen hysteresis model accurately represent the dynamic behavior of PMAs under external loading?
• RQ2 How does computed torque control based on the Bouc-Wen model compare to PID control in terms of phase response and tracking error?
• RQ3 To what extent can the proposed controller track high-frequency signals (up to 2 Hz) despite PMA-induced phase lag?
• RQ4 Does the model-based controller maintain performance under load where PID control fails?
• RQ5 Can this control strategy enable dynamic, closed-loop operation of PMA-driven continuum robots in human-centric environments?

Key findings

• The computed torque controller with the Bouc-Wen hysteresis model achieved superior tracking performance compared to PID control, especially in phase response at 0.5 Hz and 1.0 Hz.
• At 1.0 Hz, the PID controller exhibited significant phase lag, while the computed torque controller maintained better synchronization with the reference signal.
• The computed torque controller showed only about 10% overshoot but quickly corrected to follow the reference signal accurately.
• At 2.0 Hz, the computed torque controller maintained a phase lag of less than 180°, whereas the PID controller was 180° out of phase, indicating complete loss of tracking capability.
• The computed torque controller achieved acceptable tracking amplitude and phase profile at 2.0 Hz, demonstrating feasibility for high-bandwidth applications.
• The results confirm that model-based dynamic control significantly improves PMA performance under load, enabling potential use in dynamic, human-interactive soft robotics.