Abstract:

Localizing a mobile robot in indoor and GPS-denied environments is a complex problem, particularly in dynamic unstructured scenarios where traditional cameras and LIDAR-based alternative sensing and localization modalities may fail. Wireless signal-based localization has been extensively studied in the literature yet primarily focuses on fingerprinting and feature matching paradigms, requiring dedicated environment-specific offline data collection. To avoid this offline data limitation, we propose an online localization algorithm enabled by the cooperation of wireless sensor nodes (WSN) and a mobile robot that needs to be localized in an indoor environment.

The core novelty of our approach lies in that we obtain the Direction of Arrival (DOA) of wireless signals by exploiting the geometric features and cooperative communication between various sensor nodes of the WSN deployment. This is combined with a Particle Filter algorithm to calculate the Gaussian probability of wireless signal DOA online for high efficiency. The motion model of the robot (e.g., through the use of inertial or odometry sensors), if available, can be integrated into our algorithm to boost the accuracy of the localization further. The method allows online performance without needing a dedicated offline phase, as in typical fingerprint-aided algorithms.

Along with the theoretical analysis, we validate the proposed method’s accuracy and computational efficiency through extensive simulations, real-world public datasets, as well as through real robot demonstrations. The results show considerable high centimeter-level localization accuracy balanced by the high real-time computational efficiency compared to other state-of-the-art localization approaches.

Codes and Dataset: Github Link