Researchers have released a preprint detailing RL-RH-PP (Reinforcement Learning guided Rolling Horizon Prioritized Planning), described as the first framework combining reinforcement learning with search-based planning for Lifelong Multi-Agent Path Finding (MAPF), a core challenge in warehouse automation requiring multiple robots to navigate conflict-free paths continuously. The paper, posted to arXiv as preprint 2603.23838, frames dynamic priority assignment among robots as a Partially Observable Markov Decision Process (POMDP), using classical Prioritized Planning as a backbone while an attention-based neural network autoregressively decodes agent priority orders in real time. According to the abstract, evaluations in realistic warehouse simulations showed RL-RH-PP achieved the highest total throughput among tested baselines and demonstrated generalization across varying agent densities, planning horizons, and warehouse layouts. The authors report that interpretive analyses indicate the system proactively prioritizes congested agents and redirects traffic to ease bottlenecks, concluding that learning-guided approaches show potential to augment traditional heuristics in warehouse automation contexts.