This paper considers the scenario that a team of autonomous robots pursues a smart evader while concurrently building a global map in an unknown environment. We develop a modular hybrid system architecture to implement the decentralized control and coordination of multiple pursuers. A probabilistic framework is established to integrate the map building process with evader detection. Taking account of the nonholonomic constraints of vehicles, we improve the global-max pursuit policy with a novel underlying navigation method. Experimental results with physical robots demonstrate the effectiveness of our approach.