At modern warehouses, mobile robots transport packages and drop them into collection bins/chutes based on shipping destinations grouped by, e.g., the ZIP code. System throughput, measured as the number of packages sorted per unit of time, determines the efficiency of the warehouse. This research develops a scalable, high-throughput multi-robot parcel sorting solution, decomposing the task into two related processes, bin assignment and offline/online multi-robot path planning, and optimizing both. Bin assignment matches collection bins with package types to minimize traveling costs. Subsequently, robots are assigned to pick up and drop packages into assigned bins. Multiple highly effective bin assignment algorithms are proposed that can work with an arbitrary planning algorithm. We propose a decentralized path planning routine using only local information to route the robots over a carefully constructed directed road network for multi-robot path planning. Our decentralized planner, provably probabilistically deadlock-free, consistently delivers near-optimal results on par with some top-performing centralized planners while significantly reducing computation times by orders of magnitude. Extensive simulations show that our overall framework delivers promising performances.

翻译：在现代仓库中，移动机器人根据按邮政编码等分组的运输目的地，将包裹运输并投入收集料箱/滑槽。系统吞吐量（单位时间内分拣的包裹数量）决定了仓库的效率。本研究开发了一种可扩展的高吞吐量多机器人包裹分拣解决方案，将任务分解为两个相互关联的过程——料箱分配与离线/在线多机器人路径规划，并对两者进行优化。料箱分配将收集料箱与包裹类型匹配，以最小化运输成本。随后，机器人被分配拾取包裹并将其投入指定料箱。我们提出了多种高效的料箱分配算法，这些算法可与任意规划算法配合使用。我们提出了一种仅利用局部信息的分散式路径规划程序，用于在精心构建的有向道路网络上引导多机器人路径规划。我们的分散式规划器具有可证明的概率性无死锁特性，其性能稳定接近最优结果，与一些顶级集中式规划器相当，同时计算时间降低数个数量级。大量仿真表明，我们的整体框架具有令人满意的性能。