Abstract

The article considers the problem of navigating mobile robots and finding the best way to the goal in real-time in a space surrounded by unknown objects. The motor actions of the robot must be defined and adapted to changes in the environment. When using only laser scanners on mobile work, objects above or below the lasers' level will remain obstacles to the robot. Current algorithms and principles of navigation are considered. Extended the existing real-time interference detection system using lasers by adding a camera that calculates the length of objects. The new system has been successfully implemented and tested in a mobile robot, ensuring the passage of the road providing collision-free paths. The obtained simulation results are presented in the article. The existing problems of navigation of mobile robots, which are moving in the particular area from their position to the specified destination on the map, were investigated. The current problem is the inability to spot objects that are not on the same level as the mobile robot's lasers. Moreover, the task is complicated when you need to recognize such objects while the robot is moving in real time. The current algorithms and principles of navigation given by previous research and publications are analyzed. As a result of the work, the existing system of recognition and avoidance of obstacles was expanded. Prior to that, the system used only odometry and information obtained from laser scanners, without obtaining data from other sources of environmental information. The idea of development was to use a camera, which was already part of the components of the researched mobile robot. It has become possible to generate a pointcloud relative to the environment, using a depth sensing camera to calculate the distance to objects. Because the density of the received data in the form of a pointcloud is too high for further processing, a downsample VoxelGrid filter was used, which reduces the density of the point cloud. VoxelGrid belongs to the PCL library. Another problem was the removal of information about unnecessary objects in the camera's field of view. These include the floor, ceiling, parts of the robot. The PassThrough filter from the PCL library was used to solve this problem. The next step is to process the filtered data using OctoMap. As a result, an octree is generated. A top-down projection is created from the octree generated in the previous step. The resulting projection must be processed and converted into polygonal obstacles. Only then they will be marked by teb_local_planner as obstacles. The developed system was successfully implemented and tested both in the Gazebo simulation and in the researche mobile robot. The path with obstacles will be completed without collisions. The paper presents the obtained test results.