Multi-robot cooperation can significantly improve work efficiency and provide with better robustness and adaptability than a single robot. This thesis focuses on the effective cooperation strategy for multi-robot systems. Target searching in completely unknown environments is a challenging topic in multi-robot exploration. The multi-robot system has no information about the environments except the total number of targets, and a target searching task is accomplished when all the targets are detected and reached. Autonomous exploration is required to complete the task. In this thesis, a combined Option and MAXQ hierarchical reinforcement learning algorithm is firstly developed to provide with the learning ability to handle tasks in new unknown environments. Though it can work in some situations, the indispensable learning process prevents it from efficiently dealing with dynamic tasks in unknown environments. As a consequence, a potential field-based particle swarm optimization (PPSO) approach to target searching is presented. A novel potential field-based fitness function is developed for the PSO algorithm structure to provide with the exploration priority evaluation for undetected areas. The potential function is based on some designed cooperation rules. Furthermore, an improved PPSO approach with dynamic parameter tuning is developed to handle tasks in complex environments. As extensions, cooperative foraging tasks are investigated, and fuzzy obstacle avoidance is integrated to improve the smoothness of the robot path. The scheme is tested under various scenarios to validate the flexibility and effectiveness. In simulation studies, scenarios with obstacles and uncertainties are considered to demonstrate the robustness and adaptability.