With the widespread application of nuclear technology, nuclear safety becomes increasingly important. In recent years, cognitive search based on information theory has made significant progress and strongly supported the safe handling of source loss accidents, but the search efficiency needs to be improved. In this paper a multi-robot collaborative source searching method based on exploration and exploitation is proposed for obstacle environments: some robots perform the exploration, which estimates source parameter based on particle filter and get the measurements position based on the information entropy search strategy; others perform the exploitation, which calculates the global Fisher information matrix based on the real-time estimation of exploration robots to collect high-quality data to further reduce uncertainty. Simulation experiments show that by exchanging measurements, particles, estimation, a low-computation and fast search for lost radiation source in complex environments can be achieved. This method can realize the autonomous and rapid disposal of radioactive and toxic and hazardous substance leaks in nuclear power plants, chemical plants, etc.