Traditionally, an in-vehicle map consists of only one type of data, tailored for a single user function. For example, the navigation maps contain spatial information about the roads. On the other hand, a map built for adaptive cruise control use consists of the detected vehicles and their properties. In autonomous vehicle research, the maps are often built up as an occupancy grid where areas are classified as passable or impassable. Using these kinds of maps separately, however, is not enough to support the traffic safety enhancing and advanced driver assistance systems of today and tomorrow.Instead of using separate systems to handle individual safety or planning tasks, information could be stored in one shared map containing several correlated layers of information. Map information can be collected by any number of different sensor devices, and fusion algorithms can be used to enhance the quality of the information. User functions that base their decisions on the multi-layered map can then retrieve any subset of the stored information making them scalable in terms of processor and memory use.The advantages of using a shared multi-layer spatial data storage are several: Sensors and user functions are decoupled. This can make it easier and more cost efficient to implement additional functions.Data quality is enhanced. Since fusion techniques can be used to generate estimates of physical properties from several sensors, the fused data is based on all available information.Using models that describe a certain entity, properties that are not even measured can be estimated by the system.This work describes an experimental semi-autonomous ground vehicle system, where on-line generated maps containing multiple layers of information are used for obstacle avoidance and planning of a suitable path between waypoints. The system is primarily simulated using physical vehicle models in a suitable environment, but limited real-world experiments with a subset of functions are also performed.