Cooperative Relative Localization of HAVs Using UWB Technology and Machine Learning Methods in GNSS-Denied Conditions

Abstract

Automation of various technological processes often requires high-accuracy estimation of global or local object coordinates in a wide range of indoor and outdoor environments. In many practical scenarios, however, the absolute coordinate itself is less important than the relative position of an automated object, such as a robot, mobile platform, or vehicle, with respect to another object or surrounding infrastructure. Optical methods and SLAM may provide insufficient accuracy in large indoor areas with repetitive geometry. Odometry rapidly accumulates error, while the use of artificial markers is not always convenient or feasible. Even in open outdoor environments, GNSS-based navigation cannot always be considered reliable due to the increasing number of spoofing attacks. This paper presents a formal mathematical model of a cooperative localization system. The proposed approach is based on a self-organizing UWB sensor network in which navigation modules perform mutual ranging measurements and, when supported by the hardware, angle measurements. These measurements are used to form an inter-node observation matrix and to estimate coordinates in a local reference frame. Unlike conventional RTLS solutions with fixed anchor and tag roles, the proposed system treats all nodes as elements of a unified dynamic graph. To further improve accuracy by filtering NLOS measurements, an LSTM-based model is used. The proposed approach was validated through the implementation of two hardware configurations of navigation modules: with and without AoA support. Experimental evaluation showed that, in field tests, the system achieved an average deviation of approximately 7 cm relative to RTK GNSS.