Files
Abstract
Soft robots have been developed to perform delicate tasks due to the safety operation requirements. However, the compliance has brought difficulties in achieving desired motions. Although embedded sensors have been used to monitor the behaviors, they pose a risk of potential damage and reduced performance. In this article, a data-driven design strategy for soft robots is presented, utilizing finite element method (FEM) and machine learning (ML), which have provided a comprehensive understanding of soft actuators. This approach utilized massive FEM simulations, which were validated with experiments, to predict the response of actuators with different designs, generating the data for ML models. The ML model was then developed to predict the performance, allowing for the extraction of task-specific requirements to design inversely. This approach was demonstrated on a gripper, which was validated for its grasping ability with a pneumatic control system, and on a thermal metasurface, displaying a range of morphologies.