Indoor air pollution has become a growing health risk, but it is challenging to provide low-cost air quality monitoring for the indoor environment. In this article, we present “AirScope,” a mobile sensing system that employs cooperative robots to monitor the indoor air quality. Since the wireless coverage can be incomplete in some indoor areas, AirScope allows the robots to defer uploading the data to the central server by utilizing their own data buffers. In order to guarantee the timeliness of the data in the server, AirScope aims to minimize the average data latency by properly planning the routes of the robots. Such a route planning strategy has to be implemented in a distributed way since the robots that are out of wireless coverage can only make plans on their own. In addition, the cooperation of the robots is also necessary because the aggregation of the robots in a small area increases the average data latency of the other unattended areas. To solve this distributed and cooperative routing planning problem, we propose a solution based on distributed deep Q-learning (DDQL). We evaluate the system performance by simulations and real-world experiments. The results show that AirScope is effective to reduce data latency, where the proposed DDQL is 8% better than the greedy algorithm and 24% better than the random strategy.

中文翻译：

---

AirScope：通过分布式深度强化学习，移动机器人辅助协作室内空气质量传感


室内空气污染已成为日益严重的健康风险，但为室内环境提供低成本的空气质量监测具有挑战性。在本文中，我们介绍了“AirScope”，这是一种使用协作机器人来监测室内空气质量的移动传感系统。由于某些室内区域的无线覆盖可能不完整，AirScope 允许机器人利用自己的数据缓冲区推迟将数据上传到中央服务器。为了保证服务器中数据的及时性，AirScope旨在通过合理规划机器人的路线来最小化平均数据延迟。这种路线规划策略必须以分布式方式实施，因为在无线覆盖范围之外的机器人只能自己制定规划。此外，机器人的协作也是必要的，因为机器人在小区域内的聚集会增加其他无人值守区域的平均数据延迟。为了解决这个分布式协作路由规划问题，我们提出了一种基于分布式深度Q学习（DDQL）的解决方案。我们通过模拟和真实实验来评估系统性能。结果表明，AirScope 可以有效减少数据延迟，其中提出的 DDQL 比贪心算法好 8%，比随机策略好 24%。