Learning the structure and dynamics of human environments to support intelligent mobile robot behaviour

过去10年移动机器人技术的进步使在工作场所使用服务机器人(即为人类执行任务或与人类一起执行任务的机器人)变得越来越可能。对于许多应用环境，构建不随时间变化的环境地图的问题已经得到解决，使机器人能够在这些地方移动的持续时间不断增加。然而，真实的环境确实会随着时间的推移而变化，因为他们的居民会四处走动，移动家具和物品。由此导致的机器人世界的变化使其难以可靠地运行，因此存在着我们将无法创造出能够在现实情况下执行有用任务的服务机器人的危险。这项拟议的研究将机器人的环境定期变化视为一种机遇，而不是挑战。我们将开发能够从现有智能机器人(CogX项目中的探索者机器人Dora)观察到的变化中提取可靠、重要模式的系统。特别是，我们将制定两种方法。第一个将捕捉机器人在一天中的特定时间(例如，由于人类的阻挡)通过其地图的特定部分的容易或困难程度。第二种方法将捕捉物体在房间中的位置如何随着时间的推移而变化(例如，我办公室的桌子从来不动，但我的椅子往往在桌子前面移动，但从来不会靠近门)。这些方法结合在一起，将使机器人能够提高其在典型服务机器人任务中的性能，例如在建筑物中搜索物体，同时在一天中的特定时间避开特定区域(例如午餐时间食堂旁边的走廊)，所有这些都是在动态环境中。我们的研究将由空间推理和机器人学习方面的专家顾问委员会以及对使用移动机器人协助保安感兴趣的安全公司G4S提供信息。我们的结果可以帮助他们，让机器人选择更好的巡逻路线穿过建筑物，并了解房间里的物品通常是如何排列的(使它们能够发现由于入室盗窃或其他事件而发生变化的情况)。除了通过常规的科学渠道展示我们的成果外，我们还将在伯明翰的智库科学博物馆展示我们完成的机器人系统，让公众有机会更多地了解最先进的机器人和人工智能，同时也在具有挑战性的环境中测试我们的系统。

Advances in mobile robot technology over the last 10 years has made the use of service robots (i.e. robots performing tasks for, or with, humans) in the workplace increasingly possible. The problem of building maps of environments that do not change over time has been solved for many application environments, enabling robots to move around in these places for ever-increasing durations. However, real environments do change over time as their inhabitants move around and move furniture and objects as they do so. The resulting changes to a robot's world make it difficult for it to run reliably, and thus there is a danger that we will not be able to create service robots that are able to perform useful tasks in realistic situations. The proposed research treats the fact that a robot's environment changes regularly as an opportunity rather than a challenge. We will develop systems that are able to extract reliable, significant patterns from the changes observed by an existing intelligent robot (the Dora the Explorer robot from the CogX project). In particular we will develop two approaches. The first will capture how easy or difficult it is for a robot to move through particular parts of its map at particular times of day (e.g. due to humans getting in the way). The second approach will capture how the positions of objects in rooms change over time (e.g. the desk in my office never moves, but my chair tends to move around in front of the desk, but never near the door). Taken together these approaches will allow a robot to improve its performance on typical service robot tasks such as searching for an object in a building, whilst avoiding certain areas at certain times of day (e.g the corridor by a canteen during lunchtime), all in dynamic environments.Our research will be informed by an advisory board of experts in reasoning about space and robot learning, and also by the security company G4S who are interested in using mobile robots to assist security guards. Our results could help them by allowing robots to choose better patrol routes through buildings, and to learn how the objects in a room are typically arranged (allowing them to spot when things change due to a burglary or other incidents). As well as presenting our results through the usual scientific channels, we will also demonstrate our finished robot system at the Thinktank science museum in Birmingham, giving the public a chance to learn more about state-of-the-art robots and AI, whilst also testing our systems in a challenging environment.