Bluetooth RSSI-Based Lightweight Indoor Robot Localization for Edge Devices in Obstructed Environments

Abstract

Accurate indoor localization is essential for mobile manipulators to enable precise task execution and safe navigation in complex environments. Conventional GPS is unsuitable indoors because of signal attenuation, whereas vision- and acoustic-based approaches suffer from sensitivity to lighting, occlusion, noise, and reverberation. This paper presents an energy-efficient Bluetooth Low Energy (BLE) localization system that employs a lightweight multilayer perceptron (MLP) neural network designed for edge-constrained robotic platforms. Experiments are conducted on LeeAhn2, an in-house mobile manipulator, within a 3.6 m $\times 3.6$ m testbed equipped with eight Bluetooth receivers. Received Signal Strength Indicator (RSSI) samples were collected, preprocessed using filtering and smoothing techniques, and used to train the lightweight MLP for position estimation. The model maintains real-time inference capability on resource-limited hardware while capturing nonlinear relationships in RSSI data. Performance was evaluated under three scenarios: no obstacles, fixed obstacles, and unknown obstacles, demonstrating robustness against dynamic environmental changes. The system achieved a mean absolute error (MAE) of approximately 2.89% of the total range of 360 cm. By selectively excluding obstructed RSSI signals, this method ensures a high localization accuracy, resilience to interference, and practicality for deployment in energy- and resource-constrained mobile robotic systems.