LAUSR

In the soft gripper design work, most of the designs such as gripping width and the design of finger actuator are purely based on experience, and repeated trial-and-error. In most scenarios, the designed actuators cannot achieve the best/optimized grasping performance with a specific design type. This optimized design is important especially for the food grasping application, as a minor improvement of the grasping capability will be helpful to increase the grasping success ratio, especially during high-speed pick and place tasks. That motivates us to develop a design optimization framework, focusing on how to achieve an optimized grasping performance with a multi-objective design optimization. In this work, a simulation aided data-driven optimization framework for guiding the design of a reconfigurable soft gripper system is presented. To achieve an effective optimization, a simulation model is developed based on the Simulation Open Framework Architecture (SOFA) platform. This model can predict the bending and grasping behavior under actuation and external loading. This model is then used in a data-driven design optimization framework for optimizing the actuator design. An artificial neural network (ANN) is built based on the simulation results as training data, and used as a surrogate model in a multi-objective optimization framework, to achieve an optimal grasping capability with design constraints. This simulation and optimization capability can significantly reduce the trial-and-error design work, and has a great potential for effectively developing soft robots in industrial applications, such as food manufacturing and health care.