Obtaining a new type 3D LIDAR system using 2D laser scanner for autonomous vehicles and obstacle detection

In recent years, studies on autonomous vehicle technology have increased. Autonomous vehicles, also called UGV (Unmanned Ground Vehicles), are vehicles that have the ability to move on their own without the need for human intervention. These vehicles are able to detect objects in their surroundings thanks to the various sensors they carry. Obstacle distance detection, crash monitoring and reporting, accident avoidance, path planning, situation awareness, localization, mapping, and navigation are basic research topics for autonomous vehicles. To carry out the aforementioned tasks sensing systems –RADAR, IMU, GPS, LIDAR, camera, distance meter (odometer) - are used. Distance, velocity, and the angle of the objects are detected by the usage of Radio Detection and Ranging systems (RADAR), which can adapt to almost all weather conditions. However, RADAR systems cannot be used to detect the objects. IMU (Inertial Measurement Unit), one of the sensors used in autonomous vehicles, is a sensor that provides angular velocity and acceleration data to monitor the dynamically changing motion of the vehicle. It is commonly used with devices that provide location detection called GPS (Global Positioning System). While IMU provides relative position and motion information, the GPS provides absolute position and motion information. That is, when GPS determines the absolute position of the vehicle, the IMU gives information about how far the vehicle is from a position. The odometer is also used to measure the distance traveled by the vehicle. This sensor has a crucial role in the localization process. The cameras known as the eye of autonomous vehicles are used to detect obstacles, signs and traffic signals by collecting color information about surfaces within its field of view. Finally, LIDAR, one of the sensors used in autonomous vehicles and working with 3D laser scanning technique, is used to detect obstacles, visualize and obtain environmental information. These sensors, also called 3D laser scanners, have features such as high resolution, wide measurement range (up to 250 meters), detect small objects and ability to cope with reflective/difficult surfaces. At the same time, it can provide a 〖360〗^0 3D view of all obstacles around a vehicle. For this reason, LIDAR is the most suitable sensor for estimating the shape of objects according to RADAR and cameras. However, disadvantages such as high cost, being affected by weather conditions and having large dimensions prevent to their become widespread. In this study, it was aimed to develop a new type 3D LIDAR system with much lower cost, portable and 〖360〗^0 fields of view by using 2D laser scanner instead of high costing and large dimension 3D LIDARs and determine the closest obstacle with this new LIDAR system. The acquisition of 3D LIDAR is mainly based on the rotation of the 2D laser scanner around a certain axis and the fact that data of LaserScan, which is published by this laser scanner, is converted into point cloud data on a software platform. When the new LIDAR system to be obtained is examined as a whole, it needs to communicate in coordinately the components it contains in order to operate at the desired efficiency. In order to manage this software and hardware communication, ROS (Robot Operating System), which is a software platform that enables open source, free, multi-lingual, tools-based and robot components that can be controlled by the computer, is used. A ROS system consists of nodes, which performs computation, communicating with other nodes using Publisher/Subscriber messaging model. A system is comprised of many nodes and every node is responsible for one task. These nodes communicate with each other through messages published by topics. Topics carry messages that send information to other nodes. They are also called buses that used for moving messages from the node to the node. When a node sends a message via a topic, it means that the node is publishing a topic. In a similar vein, when a node receives a message via a topic, it means that the node is subscribing to a topic. Therefore, a node can publish many topics and subscribe to many topics. This study aims to obtain a more secure and stable system by integrating the algorithms created in this platform with the LIDAR system. A standard 2D laser scanner and mechanical rotational element are required for 3D scanning. In this study, HOKUYO-URG-04LX-UG01 was used as a 2D laser scanner and dc motor with encoder was used instead of a servo motor as the rotational element with a ready-made library on ROS platform. This 2D laser scanner has compact dimensions and its weight is only 160gr. It has a field of view 〖240〗^0 and an angular resolution of 〖0,352〗^0. Also, it has a maximum scanning range of 5.6m and minimum scanning range of 0.02m. This sensor is designed for indoor environment use only. Normally, the servo motor with the ready-made library on the ROS platform is subscribed to the angle data published by the ROS and ROS-Servo communication is provided. However, it is impossible to use the method applied in the ROS-Servo communication due to the angle of the ROS and the angle of the encoder do not match, which is also because of the frequency difference between the data provided by ROS and encoder. Also, the absence of an encoder library in the robot operating system is another reason why this method cannot be used. Because of the mentioned reasons, a different method has been tried in the ROS-Encoder communication. According to this method, the angle data obtained from the encoder has been sent to the ROS platform via serial communication and after the transmitted data read out from the serial, it was published in accordance with the ROS platform. So that, ROS-Encoder communication has been provided. The combination of the 2D laser scanner and the mechanical actuator has a crucial role in the system, because of the fact that different scan planes and rotation axes cause different view fields to be formed. There are four types of laser scanning methods. These scanning methods are called as pitching scan, rolling scan, yawing scan and yawing scan top. Pitching scan has a horizontal scan plane and pitch up and down. Rolling scan is rotating around the center of the laser scanner. The yawing scan and yawing scan top have a vertical scan plane and these are rotating around the upright z-axis. In this study, the yawing scan top scanning method was used. According to this method, the vertical scanning plane was taken as the scanning plane, and z-axis was selected as the rotation axis. In the mechanical design of the system was used a slip-ring component, which transfers the data on the rotating laser scanner and the dc motor with an encoder to the computer environment was used. This component prevents cables from being twisted and damaged between the rotating system and the fixed system. In addition, a test platform has been created using a laser scanner, dc motor with encoder, plexiglass table, slip-ring. Simultaneously, the L-shaped aluminum rotary support, called the Gimbal, was used in the mechanical design of the system to provide the ability to rotate around a single axis. And also coupling was used to connect the dc motor and LIDAR. Environmental information obtained from a sensor system is one of the most important steps in detecting obstacles. In order to be able to detect an obstacle, it is necessary to have information about the environment in which the obstacle is located. Thus, in this study, the 3D image of a closed environment was obtained by using the created 3D LIDAR system, so that the information about the environment was also obtained. After following these steps, the system was integrated into an autonomous golf vehicle and distance and angle of the nearest obstacle to this non-moving vehicle have been determined. Various filters have been applied to the distance and angle values determined so that LIDAR does not show the ground and autonomous vehicle as the closest vehicle. After applying these filters, the full-round scanning time of the 3D laser scanner was reduced and the position where the nearest obstacle was found was detected without error. After testing on a non-moving golf vehicle, the developed 3D LIDAR system has been integrated into a mobile robot to test 3D scanning capabilities on a moving platform. Through this robot, the system has been moved and it has been aimed to determine the position of the nearest obstacle in a moving environment. The distance and angle of the detected obstacle have been shown as the PointStamped point in the RVIZ platform, which is referred to as the ROS visualizer. The PointStamped point has been located at the center of the system when no obstacle is detected at the specified distance ranges. After the 3D point cloud image and experiments obtained by using the computer, the Jetson TX1 development card was used which provided at least as high as a computer performance. Since the Jetson TX1 is much smaller in size than the computer, it made it possible to make the 3D LIDAR system more portable. Then, a mechanical model was designed that includes the Jetson TX1 development board and all the components required for 3D point cloud viewing. Thus, the 3D LIDAR system became more portable. After the mobile vehicle used to carry the computer, a new mobile vehicle was designed to give motion to the newly created portable 3B LIDAR system and to be used in mapping, localization, and navigation. This mobile vehicle consists of 6 wheels and an aluminum frame, each driven by a separate motor. This vehicle, which works according to the differential driving method, has the dc motor with encoders attached to the medium wheels. Thanks to these encoders, it is provided to the movement control of the vehicle's wheels which based on the amount of rotation. The LIDAR platform was integrated on the designed mobile vehicle and the system was made ready for navigation. A number of algorithms and ros nodes, called Navigation Stack for navigation, are used to autonomously move the vehicle from one point to another and avoid all obstacles that may occur during movement. In order to be able to navigation with the mobile vehicle, a map of the vehicle environment was created by using the data obtained from the sensors on the vehicle. Once the map is created, the location of the vehicle on this map must be determined. Because in order to carry out an appropriate navigation, it is necessary to know which position the vehicle is in relation to the map and which direction it is looking at. So, after the mapping process, on the ROS platform performed location estimation, which is expressed as localization. Then, the path planning process was performed, taking as input the current position of the vehicle and the position it's wanted to go. Thanks to this path planning, the mobile vehicle goes to the designated spot in the fastest and best way. Finally, thanks to the navigation stack algorithm used, it was provided the vehicle is able to reach the specified target without collision to the obstacles by using odometry data (IMU and Encoder) and laser scanner data. Finally, in this study, it was used the OctoMap library, an open source framework within ROS, that provides a 3D map of the environment, specifically for C++ data structures and mapping algorithms. Thanks to the OctoMap, which can be easily integrated into many robotic systems, the 3D map of the environment was achieved using point cloud data.