Hybrid Neural Network-Based Indoor Localisation System for Mobile Robots Using CSI Data in a Robotics Simulator

Javier Ballesteros-Jerez, Jesus Martínez-Gómez, Ismael García-Varea, Luis Orozco-Barbosa, Manuel Castillo-Cara · 2025

Read Hybrid Neural Network-Based Indoor Localisation System for Mobile Robots Using CSI Data in a Robotics Simulator by Javier Ballesteros-Jerez, Jesus Martínez-Gómez, Ismael García-Varea, Luis Orozco-Barbosa, Manuel Castillo-Cara (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2511-01797.txt.

Introduction

Javier Ballesteros-Jerez and four colleagues locate indoor robots by listening to the radio channel itself. Their system infers a mobile robot's two-dimensional position from channel state information, the fine-grained signal measurements of a massive MIMO wireless setup, rather than from cameras or lidar.

The model is a hybrid neural network. A convolutional network reads channel readings converted into synthetic images by the TINTO tool. A multilayer perceptron completes the estimate. The localizer is wired into a robotics simulator through the Robot Operating System. So it can be exercised on varied test cases. It can also be combined with state estimators such as Kalman filters. The authors present the design as a general procedure, transferable beyond their specific dataset, for precise indoor navigation in complex environments where satellite positioning cannot reach.

Catalog record

CREATORJavier Ballesteros-Jerez, Jesus Martínez-Gómez, Ismael García-Varea, Luis Orozco-Barbosa, Manuel Castillo-Cara
YEAR2025
COLLECTIONPAPERS
FORMATTXT
FILE SIZE29 KB
VERSION1.0
LICENSEhttp://creativecommons.org/licenses/by/4.0/
RIGHTSCC BY
SOURCEarXiv

Source: https://arxiv.org/abs/2511-01797

Browse the full library · About the Tower of Babel

Full text

The opening of the work, from the library's hosted text:

Hybrid Neural Network-Based Indoor Localisation System for Mobile Robots Using CSI Data in a Robotics Simulator Javier Ballesteros-Jerez, Jesus Martínez-Gómez, Ismael García-Varea, Luis Orozco-Barbosa, Manuel Castillo-Cara Abstract We present a hybrid neural network model for inferring the position of mobile robots using Channel State Information (CSI) data from a Massive MIMO system. By leveraging an existing CSI dataset, our approach integrates a Convolutional Neural Network (CNN) with a Multilayer Perceptron (MLP) to form a Hybrid Neural Network (HyNN) that estimates 2D robot positions. CSI readings are converted into synthetic images using the TINTO tool. The localisation solution is integrated with a robotics simulator, and the Robot Operating System (ROS), which facilitates its evaluation through heterogeneous test cases, and the adoption of state estimators like Kalman filters. Our contributions illustrate the potential of our HyNN model in achieving precise indoor localisation and navigation for mobile robots in complex environments. The study follows, and proposes, a generalisable procedure applicable beyond the specific use case studied, making it adaptable to different scenarios and datasets. Keywords Indoor localisation \cdot Positioning \cdot Deep Learning \cdot Hybrid Neural Network \cdot Robotics Simulation. 1 Introduction Mobile robots are increasingly becoming integrated into various areas of our daily lives, including industrial tasks, medical assistance, and space exploration. Critical to their autonomous operation in these environments is the ability to accurately locate themselves. Without precise localisation, mobile robots face difficulties in basic tasks such as autonomous navigation, operational decision-making, and interaction with objects or people [ 1 ] . Therefore, knowing the exact position of the robot at each instant is essential to determine the subsequent actions to be performed. Recent advancements in robot localisation have significantly improved their autonomous capabilities, particularly through the innovative use of wireless signals. Using technologies such as Wi-Fi, Bluetooth, or ultra-wideband (UWB), researchers have developed more precise and reliable localisation systems, proposing an alternative to traditional methods of robotic localisation [ 2 , 3 ] . Wireless signals enable mobile robots to accurately determine their position in scenarios where traditional methods (e.g. GPS or visual SLAM) are ineffective. By leveraging wireless signals, robots can safely navigate complex indoor spaces, and interact with their surroundings in environments such as warehouses, hospitals, and offices. In recent years, significant advancements have been made in using Channel State Information (CSI) signals within Massive MIMO (Multiple-Input, Multiple Output) systems for indoor localisation [ 4 ] . The combination of CSI with fingerprinting techniques has emerged as a highly effective approach, leveraging the spatial diversity and multipath propagation characteristics inherent in MIMO systems. These advancements are primarily driven by the application of deep learning methodologies [ 5 ] , which allows for the direct use of signal readings to determine user locations with high precision, even centimetre-level accuracy in indoor environments [ 6 ] . Nevertheless, assessing localisation systems in real environments with these techniques presents numerous challenges, including the high costs in infrastructure, time, and human resources required for fingerprinting data collection [ 7 , 8 ] . To mitigate these problems, existing datasets containing wireless signal readings are useful knowledge sources. In addition, robotics simulators offer an effective solution for evaluating the effectiveness of localisation systems in mobile robots [ 9 ] . The goal of this paper is, therefore, to develop and evaluate robust localisation systems for mobile robots based on wireless CSI readings. Our proposal involves CSI data coming from existing datasets, which are transformed using the TINTO tool [ 10 , 11 ] to feed a hybrid neural network (HyNN) composed of a Convolutional neural network (CNN) branch with a multilayer perceptron (MLP) branch. The evaluation is performed using a robotics simulator, namely Webots 1 1 1 https://cyberbotics.com/ , which eases the design and development of different test cases. Although we have focused on the existing dataset created in [ 12 ] , this proposal holds potential for broader applications. The main contributions of this paper are: • The development of a HyNN fed from transformed CSI data, which combines a CNN branch with a MLP, and is used to perform robot localisation. • The use of robotics simulators to recreate real scenarios and integrate the localisation solution with the operation of the mobile robot. • The integration of the HyNN outputs, namely the robot position, with state estimators (e.g. Kalman filters) to analyse their value in the development of robust and accurate robot localisation systems. The rest of this paper is organised as follows: in Section 2 , we delve into the related work and review research in indoor localisation using wireless signals, deep learning-based approaches, and robotics simulation tools. Section 3 presents the proposal and analyses how to develop a localisation model from CSI readings. The experimental design is discussed in Section 4 , where we also present the obtained results. Finally, Section 5 outlines the conclusions and future work. 2 Related Work 2.1 Measurement of Wireless Signals for Indoor Localisation Recent advancements in indoor localisation employ wireless signal measurements, particularly those associated with Bluetooth, Wi-Fi, and UWB technologies [ 13 ] . Wi-Fi-based localisation is a popular method due to its widespread availability and ease of deployment, typically using Received Signal Strength (RSS). Although RSS methods are simple and cost-effective, they are susceptible to inaccuracies due to environmental variations and signal fluctuations [ 14 ] . To overcome these limitations, researchers have explored MIMO systems [ 4 ] , which can simultaneously transmit and receive multiple data streams, through the use of multiple transmission and reception antennas. This feature enhances the robustness and reliability of wireless communication, and highlight MIMO systems as a promising tool for accurate localisation. A significant advancement in indoor localisation is the use of CSI. CSI provides detailed information on signal propagation, capturing amplitude and phase information for each subcarrier, thereby enabling more precise localisation than RSS methods. Represented as a matrix of complex numbers, CSI leverages the spatial diversity of MIMO systems to create a comprehensive environment profile, thus facilitating high-accuracy indoor localisation [ 15 ] . 2.2 Fingerprinting Fingerprinting is a scene analysis technique that generates environment maps using wireless signal measurements from access points. It involves an offline phase, where reference measurements (e.g., RSS or CSI values) are collected at known locations to build a signal map, and an online phase, where real-time measurements are matched against this map to determine the device position. By leveraging the spatial and multipath propagation characteristics of CSI and MIMO systems, fingerprinting enhances the accuracy and reliability of indoor localisation. 2.3 Deep Learning for Position Estimation Deep learning has emerged as a powerful tool for enhancing indoor position estimation by leveraging the detailed signal characteristics captured by MIMO and CSI. By integrating deep learning techniques with the fingerprinting method, researchers can develop models that learn complex patterns and spatial relationships from the rich dataset of wireless signals, significantly improving the accuracy of indoor localisation systems. CNN and MLP are two types of neural networks effectively used in this context. CNNs are particularly well-suited for processing and extracting features from synthetic images, as they can learn spatial hierarchies and patterns within the data. MLPs are useful for combining these extracted features and performing precise position predictions [ 16 ]

Read the complete text