March 16, 2026 by University of Vaasa
Collected at: https://techxplore.com/news/2026-03-machine-accuracy-reliability-privacy-modern.html
While satellite navigation has become an essential part of modern life, it still struggles to work reliably indoors and in dense urban environments where high-rise buildings deteriorate signal propagation. In his doctoral dissertation at the University of Vaasa, Akpojoto Siemuri investigates how adaptive machine learning and advanced sensor fusion methods can improve positioning accuracy, robustness, and efficiency.
Accurate localization is a foundation for many essential services in modern society. It supports smartphone navigation, enables efficient logistics and transport systems, and plays a critical role in emergency response when every second matters.
However, satellite signals can be blocked by tall buildings, and often become unreliable indoors. As a result, users may experience sudden jumps in location or reduced accuracy in urban areas. Siemuri’s research in the field of automation technology investigates how machine learning can be integrated with existing sensors and smartphones to ensure accurate positioning.
“Smartphones are among the most readily available digital tools worldwide, so enhancing their positioning capability is an efficient way to support smarter cities and services. By combining global navigation satellite systems (GNSS), inertial sensors, and ultra-wideband technologies, we can achieve seamless indoor–outdoor positioning,” Siemuri says.
Trustworthy AI protects both data and battery life
The dissertation explores the use of TinyMLs, which are machine learning models that run directly on devices such as smartphones and wearables. Device-based machine learning reduces dependence on cloud-based processing, thereby strengthening privacy protection.
“By keeping the processing local, the system improves energy efficiency and contributes to discussions on trustworthy and responsible AI, including principles reflected in the EU AI Act. These models are also designed to be lightweight, which means that the improved positioning does not drain the battery or shorten its lifespan,” Siemuri explains.
Looking ahead, Siemuri explores the potential of machine learning methods for improving the orbit determination of low Earth orbit satellites, which could strengthen future positioning infrastructures.
“Because these satellites operate closer to Earth than traditional GNSS satellites, they offer potential advantages in signal strength and positioning resilience. Combined with adaptive machine learning methods, they could significantly improve localization performance in complex environments,” Siemuri explains.
More information
Akpojoto Siemuri, Adaptive Localization Using Machine Learning: Models, Methods, and Applications (2026).
Leave a Reply