Self-Organized Specialization and Controlled Emergence in Organic Computing Systems

ing from the industrial problem we study so called Emergent Sorting Networks. The basic components of Emergent Sorting Networks are router agents. Each router agent a ∈ A has an input and an output buffer with n, respectively m, positions. We denote the input buffer positions by x1, . . . , xn, and the output buffer positions by y1, . . . , ym. A buffer position can store exactly one object of k different types t1, . . . , tk. Router agents are connected to form a network topology by associating output buffer positions with input buffer positions of other agents. This is done with respect to the following conditions: (1) All input/output buffer positions must be involved in connections. (2) A one-to-one relationship between associated output and input buffer positions must hold. (3) The connections must be such that the resulting network is acyclic. Note that the set A of agents contains two special agents: the so called inflow agent serves to feed the network with incoming objects and the outflow agent produces the output of the network. Router agents can pick up objects from their input buffer positions and move them to a free output buffer position (if any). At each time step, in case the input position of the inflow agent is empty it is filled with an object of random type. Thereafter, in random order all agents apply their local routing rules. Within a time step every object can be moved at most once. The network starts empty. In Figure 2.1 the network structure that was originally proposed in Brueckner (2000) is shown. Circles represent routing agents, buffer positions are depicted as squares and arrows indicate the possible transportation direction of objects. White buffer positions