Individual deformability compensation of soft hydraulic actuators through iterative learning-based neural network

Area of Science:

• Robotics
• Control Systems
• Biomedical Engineering

Background:

• Soft actuators in robotic rehabilitation offer safe patient interaction but face control challenges due to fabrication variability and response delays.
• Existing control methods struggle with the inherent deformability and significant time lags in soft actuators, limiting their use in home-based rehabilitation.

Purpose of the Study:

• To propose a novel feed-forward control method combining a feed-forward neural network (FNN) and an iterative learning controller (ILC) for soft actuators with significant response delays.
• To enable effective learning and acquisition of the soft actuator's inverse model, accounting for individual deformability and response delay.

Main Methods:

• An iterative learning controller (ILC) was employed to iteratively learn and compensate for the soft actuator's deformability.
• A feed-forward neural network (FNN) was trained using the ILC's control results as supervised learning data to acquire the inverse model, including deformability and response delay.
• Experiments were conducted using fiber-reinforced soft bending hydraulic actuators to validate the proposed control method.

Main Results:

• The ILC successfully learned and compensated for actuator deformability.
• The iterative learning-based FNN achieved precise tracking performance across various generalized trajectories.
• The proposed method demonstrated effective acquisition of the soft actuator's inverse model, encompassing deformability and response delay.

Conclusions:

• The developed control method significantly improves the control precision of soft actuators, particularly those with large response delays.
• This approach holds promise for advancing the development of robotic rehabilitation devices and the broader field of soft robotics.
• The integration of ILC and FNN offers an efficient way to model and control complex soft actuator dynamics.