Adaptive Learning-based Predictor for Enhanced Control of Assistive Soft Robots

Area of Science:

• Robotics
• Control Systems
• Biomedical Engineering

Background:

• Soft robots offer safety and adaptability for healthcare assistive care.
• Controlling soft robots is difficult due to nonlinear dynamics and time delays.

Purpose of the Study:

• To develop a novel learning-based Smith Predictor (SP) for enhanced delay compensation in soft robots.
• To improve the control accuracy and safety of soft robots in healthcare applications, particularly for assistive care.

Main Methods:

• A learning-based Smith Predictor (SP) was proposed, utilizing Kernel Recursive Least Squares Tracker (KRLST) for online approximation of the SP correction term.
• Legendre Delay Network (LDN) was employed for efficient input history encoding.
• The predictor was integrated with a baseline robust model-based nonlinear controller.

Main Results:

• Experimental results showed significant improvements in tracking performance compared to the baseline controller.
• The proposed method demonstrated computational efficiency and online adaptability.
• The approach proved effective in enhancing delay compensation for soft robotic systems.

Conclusions:

• The developed learning-based SP offers a viable solution for overcoming control challenges in soft robots.
• The method is suitable for real-world assistive care applications, enabling safer and more accurate robot control.
• This approach has the potential to advance the use of soft robots in healthcare settings.