As automated vehicles replace the human driver, their range of tasks is immense. Most of them, like motion planning, navigation or mapping, rely heavily on the ego-position. Therefore, a robust, accurate and real-time localization system can be seen as a key requirement for assisted driving. GPS, the prevailing positioning technology, uses signals from satellites for localization. However, due to blocking and reflection, there are several environments where reliable localization via GPS is impossible, e.g. indoor environments, dense urban areas and forests. Dealing with these GPS-denied environments demands for another reliable positioning technology.

Due to recent advances in LIDAR technology, laser scanners soon will constitute a major building block for automated driving. Also localization tasks greatly benefit from the 3D nature of this sensor. Thus, this work exemplary shows the development and deployment of a localization system for assisted or automated vehicles based on range scanners. For this purpose two different algorithms are tested: Iterative Closest Point (ICP), on the one hand, is a very popular scan matching algorithm for pose tracking. Monte Carlo localization (MCL), on the other hand, applies probabilistic techniques, also known as particle filter.

Although both algorithms have very distinct working principles, they show similar performance in the experiments conducted in the FOKUS underground carpark. Both algorithms demonstrate the fulfillment of the high requirements for assisted and automated driving. Neither ICP nor MCL fail once during all 90 minutes of the test drives. In terms of accuracy, ICP outperforms MCL slightly. However, MCL proves to be the better choice in cases of great uncertainties. Besides, the MCL has the ability to (re-)localize globally. The convergence phase is passed after 40 s in average and the pose is always identified correctly. While ICP bases on a fast open source library, a novel computation method accelerates MCL to real-time for large sets of particles.