BUILDING A MAP OF REFERENCE SURFACES TO SOLVE THE PROBLEM OF PLANNING THE MOVEMENT OF A GROUP OF GROUND ROBOTS

Abstract

The purpose of the study is to form a geometric model of the environment containing information
about the parameters of the underlying surface for use in a system for planning the movements
of a group of robots in formation at high speed. The article examines the problem of con structing a map of support surfaces. An analysis of existing research on the topic of determining
the characteristics of supporting surfaces by mobile robots is presented. A classification of methods
for assessing the characteristics of a supporting surface into remote and contact ones is given.
Based on an analysis of the advantages and disadvantages of known remote and contact methods,
the work proposes a combined approach that makes it possible to use the advantages of both
methods. The approach is based on remote division of space into clusters according to the external
parameters of the underlying surface with potentially identical internal properties, simultaneous
determination of the internal parameters of the underlying surface by the contact method and their
further combination. In this case, the surface parameters are constantly updated during movement.
The approach uses a limited list of standard on-board means of a mobile robot and does not require
large computational costs compared to machine learning methods. A description is given of
the remote determination of the external parameters of the underlying surface, which are based on
point cloud segmentation algorithms that do not require preliminary training. The arguments for
segmentation are: the coordinates of the cloud points, the color of each point, and the height difference
in the vicinity of each point. An algorithm for determining the internal characteristics of a
surface using the contact method is described. The friction coefficients between each wheel and
the current surface are considered as internal parameters. These coefficients make it possible to
determine the maximum accelerations for each robot in the group, which are necessary to implement
the motion planning system. The paper presents the results of experimental studies of remote
determination of the parameters of the underlying surface within the framework of the proposed
approach using data from the public KITTI dataset. The results of the study confirm the possibility
of forming a geometric model of the environment, segmented into areas with different characteristics
of the supporting surface without training using standard hardware capabilities of the robot