Efficient Evaluation and Optimization of Automated Gripper Finger Design for Industrial Robotic Applications

Gripper fingers design is a current and important problem in industrial robotics. Recently, advances have been made to replace the arduous manual trial-and-error design process with optimization methods based on dynamic simulations. In these approaches, the gripper fingers are parametrized and evaluated by simulating multiple grasp sets in order to obtain the quality score, which is subsequently optimized. The computational efficiency of this process depends on: (1) the choice of the scoring function that provides robust evaluation with minimal number of grasps, (2) the choice of optimization algorithm that converges to global optimum quickly and (3) the choice of optimization method and meta-parameters. In this paper, we present considerations pertaining to these three problems. We use the previously proposed gripper finger design and optimization methods for generating a finger cut-out for an asymmetrical object used in industrial assembly tasks. We suggest two new alignment quality scores and compare their efficiency with preexisting methods. In addition, compare the performance of two optimization methods (one local and one global) and find the meta-parameters for the local method.