AI-based hybrid neural network with PID control system for deflection control of IPMC soft actuators in robotic applications

Abstract This paper presents a robust and adaptive real-time control system for soft actuators based on ionic polymer–metal composites (IPMCs), where key challenges are in precision control due to their non-linear behavior, hysteresis, and environmental sensitivity. The proposed novel architecture consists of an artificial neural network (ANN) with a PID controller to minimize tip deflection error effectively. An experimental test setup has been developed to collect the deflection data of the IPMC actuator. A Baumer laser sensor captures accurate deflection data, which is digitized via a MAX485 module and processed using an Arduino microcontroller. In this study, a real-world dataset has been employed to train an ANN model in MATLAB software, which facilitates accurate prediction of IPMC deflection values corresponding to given voltage inputs. The control loop is implemented in Simulink, where a setpoint deflection feeds into a PID controller, which then adjusts the voltage input to the ANN model. The maximum steady-state error is up to 0.01 mm. This novel hybrid framework compensates for system non-linearities and sensor noise, significantly enhancing actuator performance. The integration of high-resolution sensing, intelligent modeling, and closed-loop control demonstrates an efficient solution for precise IPMC actuation in grippers, manipulation, soft robotics, bio-mimetic, and other applications.