An Object-Pose Estimation Acceleration Technique for Picking Robot Applications by Using Graph-Reusing k-NN Search

Abstract

An object-pose estimation acceleration technique for picking robot applications by using hierarchical-graph-reusing k-nearest-neighbor search (k-NN) has been developed. The conventional picking robots suffered from low picking throughput due to a large amount of computation of the object-pose estimation, especially the one for k-NN search, which determines plural neighboring points for every data point. To accelerate the k-NN search, this work introduces a hierarchical graph to the object-pose estimation for the first time instead of a conventional K-D tree since the former enables simultaneous acquisition of plural neighboring points. To save generation time of the hierarchical graph, a reuse of the generated graph is also proposed. Experiments of the proposed accelerating technique using Amazon Picking Contest data sets and Arm Cortex-A53 CPU have confirmed that the object-pose estimation takes 1.1 seconds (improved by a factor of 2.6), and the entire picking process (image recognition, object-pose estimation, and motion planning) takes 2.5 seconds (improved by a factor of 1.7).