Lidar-based Robot Detection and Positioning using Machine Learning Methods

Abstract

This paper presents a machine learning-based kin detection method for multi-robotic and swarm systems. Detecting surrounding objects and distinguishing robots from these objects (kin detection) are essential in most of the multi-robotic applications. While infrared, ultrasonic, vision systems had been mainly used for applying the robot detection and relative positioning task in the literature, studies that use the Lidar-based approach is limited. The proposed method uses the Lidar sensor to discover the work area and determine the distance and the angle of all kin members relative to the observer robot. The main steps of the proposed method can be summarized as follows: 1) the Lidar distance points are read and stored as a vector with some pre-processing, 2) the acquired distance points representing different objects in the environment are separated from each other using a segmentation method, 3) in order to classify the segmented objects, the segment classification process starts with extracting five features for each object, then these features are fed to various machine learning classification algorithms to distinguish the kin robots, 4) the segments classified as a kin robot in the previous step are handled and the relative position is found for each of them. A new mobile robot prototype has been modeled and equipped with a Lidar sensor using ROS platform. Lidar has been used to collect data and four different classification methods have been tested to verify the efficiency of the method using Gazebo simulation platform.