Deep-Reinforcement-Learning-Based Autonomous Establishment of Local Positioning Systems in Unknown Indoor Environments

Local positioning systems (LPSs) serve as a feasible alternative to provide positioning service in global navigation satellite system (GNSS)-denied environments. When the area of interest is unknown and potentially dangerous, e.g., urban search and rescue (USAR), or unreachable, e.g., extraterrestrial exploration, the autonomous establishment of LPSs by a robot is an attractive approach to coping with the demand for positioning service. In this article, we investigate the autonomous establishment problem in indoor scenarios, where a robot carrying several positioning beacons intends to place them sequentially to establish high-quality positioning services for the area of interest. To solve the complicated sequential decision problem, we first model the optimal positioning beacon configuration problem and then model the autonomous establishment process as a partially observable Markov decision process (POMDP). We apply deep reinforcement learning (DRL) to solve the POMDP. Extensive simulations, including comparisons with other baselines and generalization experiments, demonstrate the advantages of the proposed DRL-based autonomous establishment of LPSs.