Embedded Six-DoF Force–Torque Sensor for Soft Robots With Learning-Based Calibration
Tannaz Torkaman, Majid Roshanfar, Javad Dargahi, Amir Hooshiar
Abstract
Soft robots typically exhibit large deformation that makes integrating rigid sensors cumbersome. Especially for soft surgical robots, direct sensor-based feedback is required. In this study, we have proposed, modeled, prototyped, and validated a novel smart polymer-based soft sensor for integration with soft robots. Previously, we have shown that the proposed smart polymer exhibits piezoresistivity. Thus, in this study, we integrated the proposed sensor with a flexural soft robot. Afterward, the sensor was calibrated through a series of experimental tests, and a multilayer perceptron (MLP) was trained for the calibration. The calibration showed a maximum root-mean-square error (RMSE) of 10.6 mN and a mean absolute error of 8 ± 10 mN compared with the ground truth. The experimental validation showed that the proposed sensor and calibration method demonstrated a combined mean absolute error of 7.4 ± 6.5 mN. In addition, the minimum detectable force of the sensor was less than 1 mN with a range of up to 284 mN. The system performance was compatible with representative intraluminal applications’ range and accuracy requirements.