<p>Autonomous collision avoidance algorithm is crucial for autonomous ships, and numerous reliable and effective outcomes have been studied. However, current research has largely overlooked the critical aspect of inter-ship interaction. In this case, autonomous ships may adopt overly cautious or tend toward active give-way, potentially compromising the efficiency of navigation and collision avoidance maneuvers. Multiple projects carried out by institutions such as the International Maritime Organization are committed to improving communication and navigation technologies for ships. It is reasonable to anticipate that advancements in equipment and conditions will increasingly facilitate ship-to-ship interaction, thereby enhancing the capabilities of ship collision avoidance systems. This paper introduces a novel collision avoidance method that leverages game theory and interaction for multi-autonomous ships at sea. Distinct from general autonomous ship collision avoidance methods, this approach places a premium on the exchange of information, advocating for a collaborative strategy. The paper outlines the challenges in coordinating collision avoidance among multiple autonomous ships, and then uses a local Self-Play strategy to simplify the complex interaction process. The method aims to make collective collision avoidance decisions through a series of interactions. Validations in different scenarios have shown that this approach could achieve collaborative collision avoidance between multi-autonomous ships.</p>