Towards Semantic Radio SLAM with Landmark Feature Extraction in mmWave Networks

Integration of radio-based sensing as part of advanced 5G and beyond mobile network standards has induced an increasing demand for high-accuracy simultaneous localization and mapping (SLAM) using cellular radio signals. Conventional SLAM solutions are built on utilization of geometric connections between the observed radio signals and the environment, and only consider estimation of landmark coordinates along with the user device state. However, besides geometric information, radio SLAM can be used to detect and exploit additional landmark features, for example, related to landmark material characteristics or the type of radio wave interaction at the landmark, conceptually referred to as semantic radio SLAM in the following. In this paper, we propose a novel method for radio SLAM with complementary landmark feature extraction, thus establishing the first steps towards the semantic radio SLAM. Based on indoor experimental 5G mmWave data at 60 GHz carrier frequency, we show that the proposed method is able to categorize landmarks into reflection and scattering objects as well as provide valuable information on the landmark material and shape characteristics. Furthermore, the utilized SLAM method achieves positioning accuracy of $20-30 \mathrm{~cm}$ and operates without any prior information on the clock bias or orientation of the user device.