A Pose Association Method for Multi-Agent System Based on LiDAR Pointcloud Registration Algorithm
Haozhen Li, Ensong Liu, Sijiang Li, Xiang Cheng · 2023
Pose association is an important problem in intelligent agents, especially in multi-agent systems, because the introduced supplementary pose information can help the localization algorithm and upper-layer modules to improve performances. This paper proposes a new method to build pose association with considerable accuracy and reliability into a multi-agent system. The proposed method utilizes LiDAR pointcloud registration to calculate the pose relationship and has good potential to be applied to most multi-agent systems as long as they are equipped with LiDAR. Modified on Iterative Closest Point (ICP) algorithm, a stepwise registration strategy based on decreased thresholds called DT-ICP is proposed. It can significantly improve the performance of pointcloud registration. Furthermore, we design an additional confidence evaluation module to avoid counter-productive pose associations caused by unreliable registration. The method has been proven to perform well in both simulation test environments and real scenarios.