PDRNN: Modular Data-driven Pedestrian Dead Reckoning on Loosely Coupled Radio- and Inertial-Signalstreams
Bauer P, Porada A, Ott F, Feigl T (2025)
Publication Language: English
Publication Type: Conference contribution, Original article
Publication year: 2025
Conference Proceedings Title: Proc. Intl. Conf. IEEE Symposium on Position Location and Navigation (PLANS)
Event location: Salt Lake City, Utah
Abstract
Classic pedestrian dead reckoning (PDR) methods struggle with dynamic motion and sensor nonlinearity due to their reliance on fixed thresholds and user-specific kinematic constraints. While recent artificial intelligence (AI)-based methods, such as RoNIN, achieve higher accuracy by modeling complex movements, they treat the entire PDR process as a monolithic black box. This approach limits control over individual processing steps and allows errors to propagate directly and unchecked to the final position. Classic methods, though general, are constrained by static thresholds, while recent AI models are often overly specific, being trained on particular individuals or scenarios. Both approaches require significant effort for calibration or optimization.
This paper introduces a modular hybrid AI-assisted PDR system, **PDRNN**, to address these limitations. PDRNN treats each component as independent ensembles of AI models, which estimate both the mean and variance for key parameters. The system employs separate AI models for orientation, velocity, and distance estimation from acceleration and gyroscope measurements, and optionally integrates absolute positions from a time-synchronized radio system, such as 5G (FR1), to initialize or stabilize the PDR system. A final fusion model combines these outputs—position, velocity, and orientation—into the final pose, with each component providing uncertainty estimates to improve robustness and resilience to errors.
PDRNN’s modular design enhances flexibility, allowing individual components to be updated, fine-tuned, or replaced without affecting the overall system. This modularity ensures more accurate and robust position estimates while avoiding error accumulation, a common issue in pure black-box AI methods. Experiments on sensor data from dynamic sports movements demonstrate that PDRNN outperforms classic and AI-based methods in accuracy, precision, and error resilience. Its adaptability to changing environments or sensor conditions, such as noise and data gaps, allows for continuous improvement without requiring a complete system overhaul.
Compared to state-of-the-art methods, PDRNN achieves higher accuracy (up to 90%), improved resilience (CEP95: PDR=1.25m, RoNIN+KF=0.46m, PDRNN=0.14m), and better forecasting performance (CEP95=0.05m at 1 s), while providing superior control over individual system components. These advantages come with the trade-off of increased system complexity, making PDRNN a powerful solution for robust and adaptable PDR systems.
Authors with CRIS profile
Involved external institutions
Germany (DE)How to cite
APA:
Bauer, P., Porada, A., Ott, F., & Feigl, T. (2025). PDRNN: Modular Data-driven Pedestrian Dead Reckoning on Loosely Coupled Radio- and Inertial-Signalstreams. In Proc. Intl. Conf. IEEE Symposium on Position Location and Navigation (PLANS). Salt Lake City, Utah.
MLA:
Bauer, Peter, et al. "PDRNN: Modular Data-driven Pedestrian Dead Reckoning on Loosely Coupled Radio- and Inertial-Signalstreams." Proceedings of the Symposium on Position Location and Navigation (PLANS), Salt Lake City, Utah 2025.
BibTeX: Download