Computational Intelligence Reports No. 032007 (February 2007) ISSN 1841 - 995X; http://www.cir.cs.ubbcluj.ro/ ©2007 Camelia Chira, Camelia-M. Pintea and D. Dumitrescu Sensitivity and S tigmergy in Agent-Based Systems Camelia Chira cchira@cs.ubbcluj.ro �   � Camelia-M. Pintea cmpintea@cs.ubbcluj.ro �   � D. Dumitrescu ddumitr@cs.ubbcluj.ro �   � Abstract A computational metaheuristic combining stigmergic behavior and gradual pheromone sensitivity of agents is proposed. Resulting multi-agent models can potentially address complex real-world problems particularly combinatorial optimization NP-hard problems. The new model employs agents able to communicate both directly and in a stigmergic manner. Stigmergic communication is based on different reactions to virtual pheromone trails produced by agents. Each stigmergic agent is endowed with a certain level of sensitivity to the pheromone allowing various types of reactions to the environment. The introduced model can be viewed as a multi-agent system composed of several sensitive stigmergic agents that can cooperate to solve complex problems. The proposed system is tested for solving the NP-hard Generalized Traveling Salesman Problem. Numerical experiments indicate the robustness and potential of the new metaheuristic. Keywords: agent communication, stigmergy, sensitivity, multi-agent system, ant colony optimization 1 Introduction Many optimization problems are NP-hard and cannot be solved within polynomial computation times. NP-hard problems arise in many and diverse domains including network design, scheduling, mathematical programming, algebra, games, language theory and program optimization. Metaheuristics are powerful strategies to efficiently find high- quality near optimal solutions within reasonable running time for problems of realistic size and complexity. The aim of this paper is to design a metaheuristic technique combining multi-agent systems, stigmergic behavior (inspired by Ant Colony Systems) and the new concept of