Hybrid-AI Grasp Planning System that Integrates Rule-Based and DNN-Based Methods for Throughput Improvement of Picking Robots

Abstract

In the logistics field, due to the declining birthrate, aging population, and shrinking workforce, there is growing demand for automation of manual handling tasks. Focusing on robotic picking operations, we developed two grasping methods for various items: rule-based grasp planning that considers the physical characteristics of the items and environment, and DNN-based grasp planning that can learn the grasping points obtained by the same method. Rule-based grasp planning is computationally time-consuming, and DNN-based grasp planning has a lower success rate. Therefore, this paper proposes hybrid-AI grasp planning that integrates these grasp planning methods. We effectively demonstrated that selecting an appropriate grasp planning method by the developed selector can improve throughput because it can combine a high success rate with fast calculation time.