Mission Planning for Long-Term Deployment using Probabilistic Environment Models

1研究内容简介，包括潜在影响随着移动机器人变得更加先进和负担得起，它们正被部署在从农田到海洋和热带雨林的各种环境中执行日益复杂的监测任务。在许多情况下，与固定传感器相比，机器人提供了更多的环境建模机会，因为它们可以覆盖更多的地面，并且不需要在环境中永久安装传感器。然而，这导致了一个问题，即机器人应该如何决定去哪里获取最有用的信息。我的研究将致力于开发提高机器人监控系统能力的算法，在农业、生态、可再生能源和许多其他领域具有潜在的应用。2目的和目标拟议工作的目的大体上是提高机器人监测系统的效率，使它们能够在其操作限制范围内提供更多有用的信息。要解决的具体目标将包括：1.设计一个可供机器人用来捕捉其环境的丰富表现的时空建模框架。部署在监控环境中的机器人必须能够以一种能够有效地为其规划系统提供信息的方式对其环境进行建模。2.通过丰富的基础环境模型展示有效的信息性规划。关于持续性建模和信息量规划的现有工作通常使用相当简单的环境模型。我们希望将这些功能扩展到上述更丰富的模型。3.综合考虑电池寿命等实际限制因素，使系统能够围绕充电需求进行主动规划。3研究方法的新颖性机器人技术近年来有了巨大的发展，而且还在继续发展，因此，对机器人监测的研究仍然提供了许多有待进一步探索的途径。在这项工作的算法基础上已经做了很多工作，但我们希望既扩展这些基础，而且-重要的-通过在真实机器人上的部署以及通过模拟测试来验证它们。4与EPSRC战略和研究领域保持一致建议的工作与EPSRC所描述的机器人研究领域非常一致。这一研究领域承认，机器人研究有可能在不久的将来促进许多主要工业部门的颠覆性技术，包括汽车、航空航天、核能、石油和天然气、农业、空间、制造、国防和建筑。这一研究领域和拟议的工作也与人工智能技术领域密切相关，人工智能技术领域的重点之一是自主和智能技术的开发。5涉及的公司或合作者拟议的工作得到亚马逊网络服务(AWS)的支持

1 Brief description of the content of the research including potential impactAs mobile robots become more advanced and affordable, they are being deployed in increasingly complex monitoring tasks in a variety of environments, from agricultural croplands to oceans and rainforests. In many cases, robots offer enhanced opportunities for environment modelling compared with fixed sensors, as they can cover more ground and remove the need for permanent installation of sensors in the environment. However, this leads to the question of how the robot should decide where to go to acquire the most useful information. My research will aim to develop algorithms that improve the capabilities of robotic monitoring systems, with potential applications in agriculture, ecology, renewable energy and many other areas. 2 Aims and objectivesThe aim of the proposed work is broadly to improve the efficiency of robotic monitoring systems, enabling them to provide more useful information within their operating constraints. Particular objectives to be addressed will include: 1. Designing a framework for spatiotemporal modelling that can be used by robots to capture a rich representation of their environments. A robot deployed in a monitoring context must be able to model its environment in a way that can usefully inform its planning system. 2. Demonstrating efficient informative planning with a rich underlying environment model. Existing work on persistent modelling and informative planning has typically used fairly simple environment models. We look to extend these capabilities to the richer models described above. 3. Incorporating practical constraints such as battery life in an integrated manner, to allow the system to proactively plan around its need to charge. 3 Novelty of the research methodologyRobotic technologies have developed dramatically in recent years and continue to do so, and as suchresearch into robotic monitoring still presents many avenues for further exploration. Much work has been done on the algorithmic foundations of this work, but we hope to both extend these foundations and - importantly - validate them through deployment on real robots, as well as through simulation testing. 4 Alignment to EPSRC strategies and research areasThe proposed work aligns strongly with the Robotics research area described by the EPSRC. Thisresearch area acknowledges the potential for robotics research to contribute in the near future to disruptive technologies in many major industrial sectors, including automotive, aerospace, nuclear,oil and gas, agriculture, space, manufacturing, defence and construction. This research area, and theproposed work, are also strongly connected with the Artificial Intelligence Technologies area,within which one of the focuses is toward development of autonomous and intelligent technologies.5 Companies or collaborators involvedThe proposed work is supported by Amazon Web Services (AWS)