Deep Reinforcement Learning (DRL): Another Perspective for Unsupervised Wireless Localization

Location is key to spatialize internet-of-things (IoT) data. However, it is\nchallenging to use low-cost IoT devices for robust unsupervised localization\n(i.e., localization without training data that have known location labels).\nThus, this paper proposes a deep reinforcement learning (DRL) based\nunsupervised wireless-localization method. The main contributions are as\nfollows. (1) This paper proposes an approach to model a continuous\nwireless-localization process as a Markov decision process (MDP) and process it\nwithin a DRL framework. (2) To alleviate the challenge of obtaining rewards\nwhen using unlabeled data (e.g., daily-life crowdsourced data), this paper\npresents a reward-setting mechanism, which extracts robust landmark data from\nunlabeled wireless received signal strengths (RSS). (3) To ease requirements\nfor model re-training when using DRL for localization, this paper uses RSS\nmeasurements together with agent location to construct DRL inputs. The proposed\nmethod was tested by using field testing data from multiple Bluetooth 5 smart\near tags in a pasture. Meanwhile, the experimental verification process\nreflected the advantages and challenges for using DRL in wireless localization.\n