TinyML and Edge AI for indoor positioning and navigation systems: a systematic review across compute scales

The absence of Global Positioning System signals indoors has spurred extensive research on indoor positioning and navigation systems (IPNSs) over the past decade. Growing applications such as robot localization, asset tracking, industrial automation, and assistance for the visually impaired drive this demand. While machine learning improves IPNS accuracy, cloud-based approaches face latency, power, and privacy challenges. Consequently, research has shifted toward Tiny Machine Learning (TinyML) and Edge Artificial Intelligence (AI), which better suit resource-constrained IPNS devices by enabling low-latency, secure, and cloud-independent operations. This paper presents a systematic review of 97 studies (2018 – August 2025) that apply TinyML and Edge AI to enhance IPNSs, focusing on resource efficiency and privacy preservation. A core contribution of this work is the systematic categorization of how TinyML and Edge AI are applied across different IPNS modalities. The review spans the full range of solutions, from lightweight ML on simple devices to advanced DL models, and examines how on-device AI enhances wireless, vision, inertial, and hybrid IPNS approaches. Comprehensive summary tables document model types, inputs, target functions, compute resources, edge platforms, and optimization methods for 97 studies, providing an evidence-based overview that maps current trends, research gaps, and deployment practices, highlighting the emerging role of specialized hardware accelerators and the limited integration of hardware-aware optimization strategies. The study highlights ongoing challenges, including resource limitations, data constraints, and limited model robustness in dynamic environments. It outlines future directions such as adaptive learning, advanced optimization, robust deployment strategies, and the development of standardized datasets and benchmarks.