Flexible pneumatic joint actuators are widely employed in medical, wearable device, and soft gripper applications due to their excellent adaptability. However, their joint angles are prone to oscillation due to material hyper elasticity and gas compressibility, thereby limiting control precision. To address this, this paper proposes a precise joint angle control method based on the CNN-AUPI (Convolutional Neural Network-Amplitude-dependent Un-parallel Prandtl-Ishlinskii) model. This approach employs the CNN-AUPI model to perform real-time fitting of the actuator’s output angle. It integrates the ISI (Iterative Search Inversion) algorithm for model inversion, enabling online adjustment of control pressure to compensate for angular errors. Experimental results demonstrate that this method achieves high-precision tracking of the joint along the desired trajectory, with a maximum mean absolute error of 0.28° under step inputs and a mean relative error not exceeding 1.59%. Compared with PI (Prandtl-Ishlinskii) and DRDPI (Deadband Rate-Dependent Prandtl-Ishlinskii) control methods, CNN-AUPI demonstrates superior performance in angular stability and control accuracy, offering an effective solution for stable angular control of flexible actuators.
Li et al. (Mon,) studied this question.