Improved performance for SLAM techniques using TRAP configured 2D LRFs

The issue covered in this thesis can be evaluated in two stages. In the first part of this work, some of the 2D laser range finder (LRF) based Simultaneous Localization and Mapping (SLAM) techniques available in Robot Operating System (ROS) have been discussed and compared with each other. These methods, which distinct differences from each other, are compared according to their advantages and disadvantages. These comparisons have been realized first in simulation environments, after then these comparisons have been done in real time experimental environment. There are two major reasons for working in a simulated environment. Among these, the battery consumption and charging problem which is the energy problem is the first reason. In real-time applications on the system, the battery drained within approximately one hour and recharged within five to six hours. This situation caused to a slowdown in the studies. It took at least six hours to wait for about an hour of work. In order to remove this problem, working in simulation environment has been preferred to solve the constraint in time. Another reason for the preference of the simulation environment is to avoid from possible damages of the equipments in any fault in the codes. In the simulation, crashing against the walls or obstacles do not damage the real system. The second objection is a hardware innovation which was proposed and the effects of this innovation were discussed. The improvement offered is a structure related with 2D LRF position configuration. In a traditionally configuration, a single 2D laser sensor is positioned at the front of the vehicle centered. In this existing structure, the mobile robot does not scan every point around itself totally, and blind spots have been occurred around the robot. However, in the proposed new 2D LRF position configuration, it is possible to perceive every point around 360° without any blind spot. This structure have been named as Transversal Positioned and is represented by TRAP. Two 2D laser sensors have been applied in the TRAP configuration and the sensors are placed at angles of 45° on the diagonal line of the mobile robot and the backs of the LRFs are facing each other. In this way, it was provided to detect everything around the vehicle without a blind spot. In the same way, the traditional structure and the TRAP structure are compared in terms of the predetermined criteria, additionally their advantages and disadvantages have been examined. ROS has been used in both real time environment and simulated environment. There are certain reasons why ROS has been chosen. One of the main reason for this selection is being an open source and freeware. Another factor is that being used by many researchers who work in academic, industrial and personal research area of robotics. Another reason for the preference of the ROS platform is the ease in which the simulation work is delivered in real time. Under normal circumstances it can be quite challenging to introduce a new hardware into the system and start using it. Considering the advantages provided by ROS, it is much easier to attach a new equipment to a platform and start to use the data from the attached equipment. It is not necessary to write code snippets separately for each piece of hardware. The existing code snippets are shared with all users because the platform is open source. The users need only change the parameters in the code fragment in the direction of the request and quickly start to introduce the hardware to the system and start to receive data. This simplicity allows the users to spent their time on code algorithm development rather than hardware attachment. One of the biggest reasons why ROS is preferred is that the codes developed for the simulation environment can be applied directly to the real-time system with minor modifications. The main focus of the thesis is the SLAM problem, which is often sought in the robotics literature. In order to be able to grasp this problem correctly, firstly, uncertainties in robotics were explained. These disclosures include filter structures used in this area, modeling mobile robot movements and sensor modeling. After the uncertainties in the field of robotics were expressed, explanations were given on localization and mapping issues. Explanations were made on what the localization and mapping problems are in the robotic research area, and what approaches are being sought. In order to understand the SLAM problem, it is necessary to understand the localization and mapping problems. The reason of that is because in SLAM these two problems have been followed simultaneously by each other. In the following, the definition of the SLAM problem is defined and information about what it is also given. ROS-compatible 2D laser sensor based SLAM methods were studied and investigated. While experimenting with SLAM techniques available in ROS three environment scenarios have been created to be compatible with the studies. These scenarios vary depending on the complexity of the environment according to the number and type of obstacles. Depending on the complexity of the environment, it is possible to name these scenarios as easy, medium and difficult, respectively. These scenarios with different difficulty levels have been proposed in order to make the comparison of SLAM methods available in ROS as fair and detailed as possible. It has been analyzed how behaviors changed in different environments. The first scenario was designed as an empty corridor consisting of only walls without any obstacles in the environment. This scenario can also be called a simple scenario. In the second scenario some static obstacles were added and the environment was tried to be more complicated. This scenario can be called the average scenario. Finally, more stable obstacles have been added than the previous one, and with this, not only are the obstacles with moving motions added. The final scenario can also be called a difficult scenario. In order to be able to perform a fair evaluation, the scenarios are fixed for each different SLAM structure, ie no changes are applied and the situation is adapted to the moving obstacles. It has also been noted that in all cases the vehicle starts at the same starting point and determines the same ending point as the target itself. Experimental studies have been carried out in accordance with these criteria. A similar comparison was made about the 2D laser sensors placed in the TRAP structure with the conventional 2D laser sensor construction. In this comparison, the same scenarios as the previous scenarios were used, and the starting points and the ending points were considered to be identical. In the third scenario, including the dynamical obstacles, the difference between the TRAP and the traditional structure was clearly visible. In a conventional 2D laser sensor structure, when a moving obstacle has been detected in the environment, the running algorithm can not directly understand whether it is a dynamical obstacle or not. Although in the algorithm in which the sensor has perceived an obstacle, this perceived structure may also be the result of a slip on the map. In situations like this the mobile robot needs additional data in order to understand what is perceived to be an obstacle or a slip on the map. In order to collect these additional data, the vehicle must rotate several times around its axis and the necessary data will be obtained as a result of these rotations. If the mobile robot encounters a moving obstacle again after it starts to move, it needs to rotate around its own axis again and get additional data. This requirement also extends the completion of the task. When the same experiment is repeated with the equipment for which the TRAP structure has been applied, the need to rotate the vehicle around its axis has been removed or reduced as far as possible. This has greatly reduced the completion time of the task. In addition, the faster completion of the algorithm also reduces the amount of battery used, thus providing a longer usage time.