Soft robots can create complicated structures and functions for rehabilitation. The posture perception of soft actuators is critical for performing closed-loop control for a precise location. It is essential to have a sensor with both soft and flexible characteristics that does not affect the movement of a soft actuator. This paper presents a novel end-to-end posture perception method that employs flexible sensors with kirigami-inspired structures and long short-term memory (LSTM) neural networks. The sensors were developed with conductive sponge materials. With one-step calibration from the sensor output, the posture of the soft actuator could be calculated by the LSTM network. The method was validated by attaching the developed sensors to a soft fiber-reinforced bending actuator. The results showed the accuracy of posture prediction of sponge sensors with three kirigami-inspired structures ranged from 0.91 to 0.97 in terms of $R^2$. The sponge sensors only generated a resistive torque value of 0.96 mNm at the maximum bending position when attached to a soft actuator, which would minimize the effect on actuator movement. The kirigami-inspired flexible sponge sensor could in future enhance soft robotic development.

In this article, a novel approach is presented for drift‐aware feature learning aimed at calibrating drift biases in soft sensors for long‐term use. The proposed method leverages an autoencoder for data preprocessing to extract expressive signal drift traces features, and incorporates drift characteristics through the latent space representation in a long short‐term memory (LSTM) regression neural network. In the results, it is demonstrated that the proposed approach outperforms other typical recurrent neural networks, such as LSTM, gated recurrent unit, and bidirectional LSTM, with a reduced root mean square error of 60% for the training dataset (≈2.5 h) and 80% for the testing dataset (≈20 h). The proposed approach has the potential to optimize the performance of soft sensors with long‐term drift and reduce the need for frequent recalibration. By compensating for sensor drift using existing prior information and limited time data, the proposed neural network can effectively reduce the complexity and computational burden of the system, without the need for additional settings or hyperparameter fine‐tuning.

This paper proposes a method for accurate 3D posture sensing of the soft actuators, which could be applied to the closed-loop control of soft robots. To achieve this, the method employs an array of miniaturized sponge resistive materials along the soft actuator, which uses long short-term memory (LSTM) neural networks to solve the end-to-end 3D posture for the soft actuators. The method takes into account the hysteresis of the soft robot and non-linear sensing signals from the flexible bending sensors. The proposed approach uses a flexible bending sensor made from a thin layer of conductive sponge material designed for posture sensing. The LSTM network is used to model the posture of the soft actuator. The effectiveness of the method has been demonstrated on a finger-size 3 degree of freedom (DOF) pneumatic bellow-shaped actuator, with nine flexible sponge resistive sensors placed on the soft actuator’s outer surface. The sensor-characterizing results show that the maximum bending torque of the sensor installed on the actuator is 4.7 Nm, which has an insignificant impact on the actuator motion based on the working space test of the actuator. Moreover, the sensors exhibit a relatively low error rate in predicting the actuator tip position, with error percentages of 0.37%, 2.38%, and 1.58% along the x-, y-, and z-axes, respectively. This work is expected to contribute to the advancement of soft robot dynamic posture perception by using thin sponge sensors and LSTM or other machine learning methods for control.

Kirigami structures, a Japanese paper-cutting art form, has been widely adopted in engineering design, including robotics, biomedicine, energy harvesting, and sensing. This study investigated the effects of slit edge notches on the mechanical properties, particularly the tensile stiffness, of 3D-printed PA12 nylon kirigami specimens. Thirty-five samples were designed with various notch sizes and shapes and printed using a commercial 3D printer with multi-jet fusion (MJF) technique. Finite element analysis (FEA) was employed to determine the mechanical properties of the samples computationally. The results showed that the stiffness of the kirigami samples is positively correlated with the number of edges in the notch shape and quadratically negatively correlated with the notch area of the samples. The mathematical relationship between the stretching tensile stiffness of the samples and their notch area was established and explained from an energy perspective. The relationship established in this study can help fine-tune the stiffness of kirigami-inspired structures without altering the primary parameters of kirigami samples. With the rapid fabrication method (e.g., 3D printing technique), the kirigami samples with suitable mechanical properties can be potentially applied to planar springs for hinge structures or energy-absorbing/harvesting structures. These findings will provide valuable insights into the development and optimization of kirigami-inspired structures for various applications in the future.

This paper proposes an autoencoder-based approach to effectively extract sensor features by leveraging an autoencoder as a data preprocessing method. The autoencoder constrains the hidden units in a bottleneck structure, resulting in a compressed knowledge representation of sensor readings. In the latent space representation, the encoded data learns and describes the most prominent latent attributes of sensor readings. The algorithm is experimentally validated in a real-world setting, demonstrating its effectiveness in accurately extracting relevant features from sensor data. Nine flexible bending sensors are utilized for posture sensing of a bellow-shaped fluidic elastomer actuator. Compared to previous studies, the results demonstrate that valuable features can be extracted without employing a large dropout rate for overfitting prevention, while maintaining prediction accuracy and reducing the entire sensor signals to half. Additionally, the training time is reduced by 7.2%. By providing a reduced and featured input to the regression neural network, the proposed approach not only prevents overfitting but also alleviates the computational redundancy and complexity brought by an increasing number of sensors.

In the distributed sensing method, various sensor arrays encompass both sensor readings and potential spatial information. While increasing the number of sensors can enhance accuracy, an alternative approach to achieve this lies in extracting spatial knowledge from combinations of horizontal and vertical sensor arrays. By adopting a scaffold-based conceptualization of sensor distribution, this study introduces a novel input pre-processing approach termed the ”Scaffold-based Pre-processing Autoencoder”(SPA) to augment distributed sensing methods. This approach leverages autoencoders to extract informative features from diverse combinations of sensor arrays arranged in vertical columns and horizontal planes. The effectiveness of the proposed approach is empirically validated in the posture sensing of a vacuum-powered bellow-shaped fluidic elastomer actuator, employing nine distributed flexible bending sensors. The results demonstrate that the neural network incorporating features extracted from sensor combinations surpasses the performance of conventional LSTM neural networks. Comparative investigations validate the distinct advantages of these arrangements, with horizontal combinations yielding improved estimation accuracy in the X and Y axes, and vertical combinations enhancing accuracy in the Z axis. Collectively, these combinations yield reductions in RMSE of 9.51mm to 6.93mm in the X-axis, 6.77mm to 5.43mm in the Y-axis, and 7.72mm to 4.38mm in the Z-axis. Besides, the application of SPA to GRU and BiLSTM models also demonstrated significant improvements. Both models exhibited a substantial reduction in the sum of RMSE in each axis, with decreases of 20% and 27%, respectively.