Improving Local Navigation by Application of Scan Matching Techniques in Mobile Robotics

Kurzfassung

This thesis is about the use of registration algorithms for navigation of a mobile robot. Generally, the use of these algorithms for the localization and a modification of the navigation system are examined. The Lib3D framework of the Robotics and Machatronics Center of the DLR serves as experimental evaluation environment. The main goal here is to improve the navigation through narrow environments. In the current state the robot often corrects its position a few times until it drives through the narrow environment. Through a more accurate pose estimation the uncertainty before driving through a narrow environment should be lowered. So the robot can try alternative ways to not move and relocalize again. The used registration algorithms have already been implemented in the Lib3D framework. A concept is created that depicts how the algorithms should be applied and evaluated, in order to find the best one, which can be integrated into Lib3D. This concept serves as a rough plan for the implementation. Soon in this implementation it turns out that the ICP is the best choice for the integration. It is integrated into the navigation system of Lib3D and an exact evaluation is performed. On the one hand the ICP is mostly more accurate, on the other hand the results vary strongly, what makes them less meaningful. The Conclusion is that a good statement about whether the ICP should be used or not can not be made at this point. Further examinations are necessary.