Swarm algorithm development for automated warehouse solutions

已经有大量关于群体机器人的研究表明，群体算法如何使用简单的规则、分散的控制和一组机器人来解决复杂问题。然而，关于将蜂群系统过渡到实验室之外的研究一直很缺乏。已经提出了许多可能的现实世界应用领域，如搜索和救援[1]、建筑[2]或太空探索[3]。缩小现实差距将需要新颖的实现方法，包括人与群的互动和考虑用户对群的看法。该项目旨在消除对具有自动化储存和检索系统(AS/RS)的仓库中复杂的现有基础设施或编程的需要。如果没有这些昂贵和复杂的功能，AS/RS可以在目前无法使用的领域实施，例如中小企业、慈善捐款仓库或高街商店。它们的空间通常有限，需求波动很大，这使得库存很难有效地操纵和组织。他们可用于库存分拣的劳动力也很有限，这意味着他们的系统可能效率低下、速度慢且不准确。群体机器人可以：向上和向下扩展到不同的空间；适应不断变化的任务；以及对单个代理的故障具有健壮性。在任务过程中，群体可以在没有任何特定的预编程或人类助手的干扰的情况下做到这一点。集群仓库也是一种潜在的比传统AS/RS更便宜的解决方案，因为单个代理在编程和技术上都不那么复杂(只要它与拥有多个机器人的硬件成本相平衡)。凭借这些优势，集群自动化立体仓库有可能执行与目前为提高自动化仓库效率而实施的任务类似的任务，但需要的专用控制和基础设施较少。预计这将扩大该技术在劳动力有限但库房类型、任务和规模方面存在较大差异的应用领域。本项目的工作将主要集中在开发和设计集群仓库AS/RS的算法。这将通过对这项技术的潜在用户的采访来了解，其中将包括对工作场所蜂拥而至的公众舆论的调查。该项目的最后部分将使用东芝的集群试验台对集群算法进行物理测试。一个成功的算法将在所描述的存储和检索任务中执行与具有全局信息的集中式系统相同或更好的性能。

There has been plenty of research about swarm robotics that shows how useful swarm algorithms can solve complex problems using simple rules, decentralised control and a group of robots. However, there has been a lack of research into transitioning swarming systems out of the laboratory. Many possible real-world application areas have been suggested such as search and rescue [1], construction [2] or space exploration [3]. Closing the reality gap will require novel implementation methods including human-swarm interaction and consideration of how users think of swarms. This project aims to remove the need for complex pre-existing infrastructure or programming in warehouses with automated storage and retrieval systems (AS/RS). Without these expensive and complicated features, an AS/RS could be implemented in areas where it is currently unavailable such as SMEs, charitable donations storage or high street shops. These are typically limited in space and have large fluctuations in demand making the stock difficult to manipulate and organise efficiently. They also have a limited workforce available for stock sorting which means their systems can be inefficient, slow and inaccurate. Swarm robotics can: scale up and down to different spaces; adapt to changing tasks; and be robust against single agent failures. The swarm can do this without any specific pre-programming or interference from human helpers during the task. A swarm warehouse is also a potentially cheaper solution than a classic AS/RS because the individual agents are less complex in both programming and technology (as long as it is balanced with the hardware cost of having multiple robots). With these advantages, a swarm AS/RS has the potential to perform similar tasks to the ones that are currently implemented to improve efficiency in automated warehouses but need less dedicated control and infrastructure. This is expected to increase the application areas for the technology where the workforce is limited but there is a large difference in stock room types, tasks and sizes.The work for this project will mainly be focused on developing and designing algorithms for a swarm warehouse AS/RS. This will be informed by interviews with potential users of the technology which will include an investigation into public opinion of swarms in the workplace. The final part of the project will be physical testing of the swarming algorithms, using the Toshiba swarm test-bed. A successful algorithm will do the same or better performance than a centralised system with global information at the storage and retrieval tasks described.