Akademik Veri Yönetim Sistemi
IEEE Access, 2026 (SCI-Expanded, Scopus)
Indoor positioning is a critical component of navigation and control in automated guided vehicles (AGVs) and autonomous mobile robots (AMRs). Wi-Fi RSSI–based localization provides a cost-effective solution for indoor environments; however, its deployment on edge devices is constrained by signal instability, latency requirements, and limited computational resources. This study proposes a two-stage RSSI-based localization framework specifically designed for TinyML enabled embedded systems. In the first stage, a lightweight quantized neural network generates a rapid coarse position estimate with deterministic execution on microcontroller hardware. In the second stage, localization accuracy is refined within a constrained spatial window using dual axis-based TinyML models that enforce mutual geometric consistency. The influence of access point placement is systematically evaluated using square, triangular, hexagonal, and random topologies with trajectory-based data collection. Experimental results demonstrate that random and hexagonal layouts provide superior localization stability and accuracy. Both dataset-based evaluations and real-world measured RSSI samples confirm that the fine localization stage significantly reduces localization error while preserving real-time operability. In addition to accuracy analysis, the proposed localization pipeline is implemented using fixed-point TinyML inference and executed on microcontroller-class edge hardware. The results demonstrate that the system maintains low latency, bounded computational complexity, and minimal memory usage while operating in real time. The proposed framework therefore provides an effective balance between accuracy, robustness, and computational efficiency for embedded autonomous applications.