To address the issue of multi-robot source seeking in obstacle environments, which often leads to local optima and inefficient pollution source localization, an improved Dynamic Role Allocation-Ant Colony Optimization (DRA-ACO) algorithm is proposed. Firstly, the heuristic function of the proposed algorithm integrates environmental features such as concentration gradient, spatial distance, and wind direction, enhancing the robot"s autonomous decision-making ability. Secondly, global search capability is improved by introducing adaptive pheromone volatilization and a visit frequency penalty mechanism. Meanwhile, an adaptive balance between exploration and exploitation is achieved through the dynamic role allocation strategy. The simulation results show that a significant enhancement in the source-seeking success rate is exhibited by the DRA-ACO algorithm in obstacle scenarios, compared with the traditional Ant Colony Optimization (ACO) and the Hybrid teaching learning particle swarm optimization (HILPSO). Furthermore, through simulation comparisons with gradient-based source localization, probabilistic source localization, and biologically-inspired source localization methods, the proposed DRA-ACO algorithm is verified to have stronger effectiveness and robustness in multi-robot cooperative source localization under different environments.