Design Optimization of The Directly 3D-Printed Jamming Structure Using Machine Learning Models

Modifying the stiffness of soft robots presents a significant challenge, with the goal of enabling free deformations at low stiffness, while increasing the robustness of grasping at high stiffness values. Among the promising approaches, the utilization of jamming structure stands out for its ability to achieve high stiffness modification with minimal volume variation. However, most of the jamming structures require a complicated fabrication process, which limits the prototyping speed, repeatability, and reliability. To address this issue, attempts have been made to directly 3D-print jamming structures; however, these attempts were limited to 3D-printing the standardized jamming structures and the optimized designs relied on heuristic searches based on trial-and-error experiments. In this study, we introduce a directly 3D-printed jamming structure with optimized design through machine learning. The stiffness of the generated jamming structure design is estimated by finite-element-based simulations and the design is optimized using machine learning models. To validate the optimization results, benchmark designs and the optimized design are fabricated and the stiffness of the designs are compared. Finally, the feasibility of the optimized 3D-printed jamming structure is verified by grasping demonstrations, including its implementation into a pneumatically actuated soft gripper.