Hybrid PI-Neural Network Control for Fluid-driven Origami-inspired Artificial Muscle

Pneumatic artificial muscles (PAMs) have been introduced as actuators due to their low weight, low mass-toforce ratio, compliance, high prevalence in nature and ability to closely mimic the functions of human biological muscles. When PAMs are applied in robotics and rehabilitation applications, it is essential that the actuator is operated according to user requirements. PAMs, however, present significant challenges in modeling and control due to their time-varying parameters, complex hysteresis, and highly nonlinear properties. This paper proposes an approach for controlling the motion of a PAM. This method applies a hybrid control algorithm to control a fluiddriven origami-inspired artificial muscle (FOAM). By combining a PI controller with feed-forward neural network control, the controller can learn and adapt through the system's behavior. The control algorithm was tested to observe the performance of the controller for displacement control of FOAM via different signals. Additionally, experiments were conducted to evaluate its performance under different load conditions. The results demonstrate exceptional controllability, even when the system faces increased loads, demonstrating the adaptability of the controller to load variations.