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

TL;DR

The paper tackles indoor localisation for mobile robots using CSI data from Massive MIMO and introduces a Hybrid Neural Network (HyNN) that fuses CNN-derived features from CSI-generated images with an MLP operating on CSI features to predict 2D positions. It implements a full simulation-aware workflow, generating synthetic CSI images with TINTO, integrating the HyNN within a ROS/Webots pipeline, and pairing it with a Kalman filter for state estimation. Key contributions include the HyNN architecture, a simulator-based evaluation pipeline, and a generalisable procedure adaptable to other CSI datasets and scenarios. Results show that with 32–64 antennas the system achieves mean positioning errors around $0.10\ \text{m}$ or better, with Kalman filtering providing robustness in noisy conditions while highlighting limitations in certain nonlinear or high-accuracy settings. This work demonstrates a practical, cost-effective path toward real-world indoor robot localisation using existing CSI data and simulation tools, with clear avenues for real-world validation and alternative estimators such as Monte Carlo localisation.

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.

Figures (7)

• Figure 1: Simplified architecture of the Hybrid Neural Network.
• Figure 2: Scenario configuration. The antennas are spaced around a scenario where yellow zones represent CSI readings positions. All measurements in the figure are in millimetres (Figure obtained from dataset).
• Figure 3: Reconstruction of the scenario in Webots.
• Figure 4: Box plot of the prediction times in seconds of the models as a function of the number of antennas.
• Figure 5: Results of predictions obtained in Experiment 1.