Abstract: In a complex and unknown battlefield environment, unmanned vehicle clusters can perform more complex tasks than a single unmanned ground vehicle (UGV). The formation and obstacle avoidance of unmanned vehicle clusters is one of the research hotspots in the field of swarm intelligence. To reduce the deviation between the actual formation position and expected formation when unmanned vehicle clusters avoid obstacles in an unknown environment, this paper proposes a cooperative formation and obstacle avoidance control method for unmanned vehicle clusters based on an improved dynamic window approach (DWA). In addition to the azimuth evaluation, obstacle evaluation, and speed evaluation factors of the path evaluation function in DWA, the direction coordination and formation maintenance factors are added. We established a mathematical model of unmanned vehicle cooperative formation and calculated in real time the expected position, direction, and speed of each following vehicle according to the expected formation. The sum of the deviation between the real driving and expected directions of each following vehicle was taken as the direction cooperation factor, and the sum of the absolute deviation between the real and expected positions of each following vehicle was taken as the formation-keeping factor. When unmanned vehicles approach unknown obstacles, the adaptive collaborative adjustment of the relative position and speed is performed based on the improved DWA, which can improve the precision of the formation positions during the obstacle avoidance process. In this paper, a simulation is conducted using a scenario of three unmanned vehicles forming a triangular formation to avoid obstacles and three obstacles. The simulation results show that the control adjustment based on the improved dynamic window and variable weight optimization algorithm is more timely than the traditional formation control based on the artificial potential field. Additionally, the formation position deviation is relatively small when avoiding obstacles. The average deviation between the actual position of the vehicle formation and the desired position of the formation near the obstacle is used as the evaluation function. The coefficients of the path evaluation function of the improved DWA are optimized based on the variable weight pigeon-inspired optimization (PIO). It can improve the formation, maintaining accuracy in obstacle avoidance. In summary, the proposed improved algorithm can make the unmanned vehicle cluster adjust the driving speed and direction cooperatively in obstacle avoidance. Moreover, it can improve the stability and accuracy of the dynamic formation.