Indoor Intelligent IoT Datasets for Behavioral Learning and Autonomous Decision Making

The current indoor Internet of Things (IoT) system is mainly realized by presetting rules, voice interaction, or remote control, which fails to meet the need for systematic, global, and autonomous intelligence. The cost of high-quality large-scale data annotation is also too high, which hinders the effective application of supervised learning paradigm in such scenarios. Therefore, unsupervised or reinforcement learning is worthy of consideration for the future IoT system. In view of the large-scale organization and deployment, high dynamic coordination, and precise service requirement of the future IoT system, this study constructs a perceptual data acquisition platform to obtain the indoor intelligent IoT datasets suitable for unsupervised or reinforcement learning. First, various environment-aware and controllable electrical devices are deployed in a multi-person office area that is divided into multiple zones. The devices are connected to the edge gateway through long range radio (LoRa) wireless network, and an message queuing telemetry transport (MQTT) cloud server is constructed to realize the design of cloud edge collaborative architecture. Then, two kinds of information, namely, environment and controlled action, are collected in real time without human perception, which forms and publicizes complete datasets, including time, space, environmental parameters, and control actions. On the basis of the datasets, relevant data analysis is conducted and preliminary experimental results using unsupervised learning method are presented.