Position and orientation of individual nodes in ad hoc sensor networks are useful for both service and application implementation. Services that can beenabled by availability of position include routing andqueiying.Atapplicationlevel, position is required inorder to label the reported data i n a sensor network, whereas position and orientation enable tracking. Nodes may have local capabilities such as the possibility of measuring ranges to nejghbors, angle of arrival, or global capabilities, such a s GPS and digital compasses. This article surveys methods used to infer locations in a multihop fashion in networks with or without the mentioned capabilities.ith the ever decreasing cost and size of sensors, t h e instrumentation of the world becomes possible, with a wide range of meteorological, commercial. and Dersonal applicatibns. Sensor network;, when not micromanaged in terms of topology, are actually ad hoc nctworks, in most cases not mobile o r with occasional mobility. What they share with mobile a d hoc networks is the probabilistic nature of the graph, the problems of connectivity and density control, medium sharing, and scalability. They may not face the mobility problem, but they have a host of new ones. Sensor nodes have much lower processing capabilities than current cell phones or ' PDAs; for very large scales of deployment, energy consumption is a factor that affects the trade:off between node power and thc cost of replacing batteries; configuration and maintenance on a per node hasis are not possible. One lesson that has becn learned from the work on more general, and possibly mobile, ad hoe networks is that solutions from the wired world do not usually apply directly to large ad hoc topologies, If we only look at the example of routing, link state is out of the question if we deal with hundreds of thousands of weak nodes, whereas distance vector, although more scalable, has disadvantages with respect to mobility. Both link -state and distance vector try to infer data about the entire topology at each node. This makes them undesirahle for very large ad hoc networks. At the service level, availability of position enables implementation of algorithms with better scalability. Position-centric addressing was first proposed i n t h e 1970s, but has regained attention recently with the advent of very large networks, such as sensor networks. Here, nodes are named by their position in Euclidean space, and only one node can have a given position. From this bijective relation, it follows that there is no separate job to be performed to support routing. This means that the current state of the packct (e.g., its position) and the position of the destination are enough to determine the process of forwarding. T h c simplest example of position-centric addressing and routing is Cartesian routing [2], which greedily decides that the next node to receive the packet is the neighbor gcographically closest to the destination. While this may sound simple, there are in fact several possible strategies for positio...
