UPCommons. Portal del coneixement obert de la UPC
Publication
Artículo científico (JournalArticle)
Deep learning-based UWB-IMU data fusion for indoor positioning in industrial scenario
Accurate and precise wireless infrastructure-based positioning systems become crucial as industries move towards flexible, portable, and autonomous transportation systems such as Automated Guided Vehicles (AGVs). Multipath-dominant dynamic environments like industries present significant challenges for wireless signal propagation and affect wireless positioning accuracy due to the interplay of reflected signals from obstacles. The achievable indoor positioning accuracy of the target AGV can be enhanced by fusing the measurements from the wireless infrastructure with the target's onboard sensor data. Nevertheless, the lack of correspondence between the wireless infrastructure and the target's onboard sensors causes the measurements from these two systems to arrive at irregular time steps. Using asynchronous measurements in the data fusion process can degrade the overall positioning accuracy of the target AGV. This paper proposes a novel deep learning-based data fusion approach to deal with asynchronous measurements from the wireless infrastructure Ultra-Wideband (UWB) and the target's onboard Inertial Measurement Unit (IMU) sensor to achieve enhanced positioning accuracy of the target AGV. In particular, a two-stage cascaded Deep Neural Network (DNN) is proposed to deal with the asynchronized measurements from UWB and IMU sensors. The first stage of the DNN is used to obtain the initial position estimate of the AGV by processing the measurements from UWB. Subsequently, the second stage of the DNN fuses the initial position estimate of the AGV with the IMU sensor data to obtain the final enhanced position estimate. The proposed approach is validated with real-world experiments in an indoor industrial scenario using UWB technology in channel 2 (3.7-4.2 GHz) and an IMU sensor placed on an AGV. Moreover, the achievable positioning accuracy and the computational runtime to provide the position estimates with the proposed approach are analyzed. The experimental results show that the proposed approach achieves a mean absolute error of less than 10 cm, outperforming the considered baseline methods, Extended Kalman Filter (EKF) and Long Short-Term Memory (LSTM)., The work of Josep Paradells was supported in part by the Spanish MCIU/AEI/10.13039/501100011033/FEDER/UE through Project PID2023-146378NB-I00 and in part by the Secretaria d’Universitats i Recerca del departament d’Empresa i Coneixement de la Generalitat de Catalunya under Grant 2021 SGR 00330. This work was supported in part by the European Union’s Horizon 2020 Research and Innovation Programme through the Marie Sklodowska-Curie under Grant 956670, in part by Project 6-SENSES under Grant PID2022-138648OB-I00 funded by MCIN/AEI/10.13039/501100011033, and in part by ERDF A way of making Europe., Peer Reviewed
Project:
Agencia Estatal de Investigación / Plan Estatal de Investigación Científica y Técnica y de Innovación 2021-2023 PID2023-146378NB-I00
,
European Commission / Horizon 2020 Framework Programme 956670
,
Agencia Estatal de Investigación / Plan Estatal de Investigación Científica y Técnica y de Innovación 2021-2023 PID2022-138648OB-I00