The aim of landmark-based navigation is to reduce the vehicle's dependence on satellite-based navigation systems such as GPS. Instead, the car is to store perceived landmarks on a map using the on-board sensors. If the same or another vehicle drives the same route again, it can now use the perceived landmarks to locate itself on the existing map and navigate in a targeted manner without knowing its exact global position. Landmarks are mainly objects that a human being would also use for orientation: intersections, streets and static objects that stand out from their surroundings, such as houses or individual trees on a meadow. However, abstract features can also be used in camera images that are extracted from a neural network, for example.

Currently, the g2o library is used to create a metric-topological map, the Levenberg-Marquardt implementation of which optimizes the map taking perception uncertainties into account. A particle filter is then used for localization within the map. Once the autonomous vehicle has successfully located itself, the path that the car followed when creating the map is transferred as the desired route to the trajectory planning module, which guides the vehicle to its destination while avoiding obstacles.

Due to the efficient internal representation of the landmarks used here, these can also be transmitted by radio to other vehicles. This functionality is used in the convoy scenario when the line of sight between the following and command vehicle breaks off. Using the landmark data received from the command vehicle, the following vehicle maps the route of the command vehicle and locates itself on the resulting map. Thus, even without visual contact, it is able to follow the path of the command vehicle and at best catch up to it again in order to continue the regular convoy journey.

In a further step, the previously metric-topological map is to be reduced to purely topological information. The registered landmarks then merely represent a kind of human-readable route description with which the autonomous vehicle can reach its destination without knowing how many metres it still has to cover. Such a description could take the following form: "Drive along the dirt road to the third intersection, turn right there and stay at the house painted red". The challenges here lie primarily in the precise recognition and classification of landmarks.