End Positioning Accuracy Improvement for Soft Arms Under Various Loads

Abstract

The multicavity pneumatic soft arm is usually controlled based on an open-control system. For various loads that act on the end, the arm would generate significant deviations under the constant air pressure input to the cavities. To improve the accuracy of end positioning, this study developed a flexible strain sensor made of liquid alloy, which is embedded into the soft body of the arm. In addition, a kinematic mapping is established using a deep learning neural network to overcome the influence of the nonlinear and complicated deformation of the soft body. By incorporating the resistance feedback from the sensors, the model can accurately calculate the end position of the arm. The end positioning accuracy and trajectory tracking performance of a two-degree-of-freedom (2-DOF) pneumatic soft arm based on the proposed sensors and the mapping were experimentally measured. The results demonstrated that the maximum error of the end position was 5 mm under various loads. Therefore, the proposed method can be applied to a wide range of soft robotics and has the potential to improve the accuracy and robustness of their control.