The COMPASS Location System Frank Kargl and Alexander Bernauer University of Ulm, Dep. of Multimedia Computing, Ulm, Germany Abstract. The aim of COMPASS (short for COM mon Positioning Architecture for S everal S ensors) is to realize a location infrastructure which can make use of a multitude of different sensors and combine their output in a meaningful way to produce a so called Probability Distri- bution Function (PDF) that describes the location of a user or device as coordinates and corresponding location probabilities. Furthermore, COMPASS includes a so called translator service, i.e. a build-in com- ponent that translates PDFs (or coordinates) to meaningful location identifiers like building names and/or room numbers. This paper gives a short overview on the goals and abilities of COMPASS. 1 Motivation There are a lot of situations in mobile computing where mobile nodes need to determine their current position. Ubiquitous computing applications derive context information from this position, e.g. in order to determine whether a user is currently at home, at work or on the way in between. Location-aided routing protocols for ad-hoc networks need position information to support their routing decisions. Navigation systems naturally rely on precise position information to plan the further route of a car or pedestrian. To support this large demand that applications have for precise location information, a number of commercial and research projects are working on this subject. Section 2 gives an overview on some of these activities. We have identified two major challenges that are not completely resolved yet: 1. Location information from multiple sensors needs to be combined effectively in order to present one and only one position to the application. Any single location sensor has drawbacks, e.g. is usually not available inside buildings, RFID sensors or WLAN/Bluetooth APs are only available where installed etc. So in order to provide reliable and pervasive location support, an ar- chitecture must use multiple sensors, combine their results and present this to the application. The application should not need to worry about what sensor(s) were used for the current position information. Additionally com- bining the results from multiple sensors may improve the precision of overall results. 2. Raw coordinates may not really be useful to an application that needs to know the position in terms of buildings, rooms, street names etc. So a lo- cation system should include an infrastructure to resolve the raw position information to some kind of symbolic position. T. Strang and C. Linnhoff-Popien (Eds.): LoCA 2005, LNCS 3479, pp. 105���112, 2005. c Springer-Verlag Berlin Heidelberg 2005

106 F. Kargl and A. Bernauer The primary focus of COMPASS will be to address these two issues by both including many different sensors into the system using a plugin interface and by providing a translator that is able to derive symbolic location information from the raw coordinates received from the locator. COMPASS is a software framework that can be used by arbitrary applications for location retrieval. 2 Related Work The need of location systems is almost as old as mobile computing itself. Many of them use satellite navigation systems like GPS [G93] or the future Galileo system [G05]. A major problem of satellite navigation systems is the fact that the antenna of the receiver usually needs a direct line-of-sight towards a number of different satellites. So they are only useful for outdoor navigation. As many ubiquitous computing projects include mostly indoor scenarios, re- searchers started to develop specialized indoor location systems. Prominent ex- amples include the Cricket Location-Support System [P00] or the Bat [H97]. These systems make use of different kinds of sensors, like scanning for ultra- sound or radio beacons or observing nearby WLAN or Bluetooth access points. Unfortunately most of these location systems do not work together and many can use only one single kind of sensor. So there is a clear need for a framework that can combine the output of different kinds of location sensors into one single and consistent result. Such a system has been proposed as part of the HeyWow project [H03]. In [A01, W02] the authors suggest the use of so-called probability density func- tions (PDFs) to represent the location measurement of one sensor or the com- bined measurement of multiple sensors. Other similar projects include [B03, H02]. As our architecture is based in part on these ideas, we first give some details on how position is represented in COMPASS before describing the architecture itself. 3 Position Representation A major issue in positioning systems is how to express the position. COMPASS knows two kind of position representations: a geocoordinate based representa- tion and one that delivers a semantical description of the current position, like the current room number or a street address. The functionality of COMPASS includes a mechanism to translate a geocoordinate to a semantic position de- scription automatically, as sensors often deliver the first representation whereas applications often need the semantic representation. No matter what kinds of sensors are in use to determine geocoordinates, most of them will inevitably introduces some kind of error. E.g. GPS has a typical error of a few meters, estimating the position based on available WLAN access points will deliver results with a precision of a few dozens to a few hundreds of