In large-scale industrial environments where multi-AMR (Autonomous Mobile Robot) systems are deployed, the unpredictable occurrence of obstacles can significantly disrupt AMR navigation, hindering task execution. To overcome such disruptions, AMRs must frequently replan their routes in real time, often resulting in suboptimal trajectories. This paper proposes a multi-AMR path planning framework based on a Cooperative Costmap with Lifelong Learning, designed to enable efficient navigation even in environments where obstacle patterns are not known a priori. Inspired by issue-propagation models in social-network theory - which describe how public attention rises and fades over time within a network - the proposed approach models the temporal influence of encountered obstacles, allowing predictive path planning that adapts to changing obstacle patterns. The framework incorporates a lifelong learning mechanism to incrementally refine the influence parameter over time, thus ensuring adaptability in dynamic industrial settings. Simulation experiments demonstrate that the proposed approach increases task throughput by up to 18.0% and reduces average travel time by up to 30.1% compared to the standard ROS 2 navigation stack.
