A Data-Driven Topology Optimization Framework for Designing Robotic Grippers

A widespread methodology to enhance the design of robotic devices is represented by topology optimization. Typically, the optimization aims at designing a certain part of the robot to satisfy a priori, user-defined mechanical properties while minimizing the used material for building the structure. In this paper, we apply topology optimization to robotic grippers, and we propose to define the requirements for the optimization in a data-driven way based on simulated experiments of grasping tasks. Specifically, the architecture we propose is composed of three sequential phases. The input of the architecture includes the initial model of the gripper, the specific gripper component to be optimized, and a set of parameters. The first part of the architecture acquires force signals from the gripper component that are sensed during the grasping simulations. Hence, these signals are fed into the second phase, which analyzes the forces through pixel connectivity and Dynamic Time Warping algorithms and provides the instructions for the topology optimization. Ultimately, the third block performs the optimization. The method is tested by optimizing a specific part of a soft-rigid gripper. Results from simulation confirm that the proposed architecture provides an improved version of the original gripper, not only in terms of optimized use of materials but also in terms of grasp success rate.