Description

Precise indoor localization using radio signals exchanged between base stations (BSs) and mobile “agents” (radio devices such as mobile phones) remains a challenging problem for several essential applications such as search-and-rescue and autonomous navigation. Simultaneous localization and mapping (SLAM) generates a feature map of the environment and localize the agents within that map in such situations. One promising solution is radio frequency (RF)-SLAM where specular reflections of radio signals off flat surfaces are used to estimate the locations of both an mobile agent and the features representing reflecting surfaces. Model-based state-of-the-art (SOTA) techniques for RF-SLAM formulate and solve a high-dimensional sequential Bayesian estimation problem using factor graphs [1, 2]. All these methods have in common that their robustness and accuracy depends strongly on chosen feature models and how well these features represent the geometry of the environment. If a significant model-mismatch oocurs and the feature cannot be well extracted from the data, these algorithms have the dendency to fail in reconstructing good feature map. To overcome these issues, we propose a deep learning approach in line with [3] to overcome these issues.

References

• [1] E. Leitinger, F. Meyer, F. Hlawatsch, K. Witrisal, F. Tufvesson, and M. Z. Win, “A belief propagation algorithm for multipath-based SLAM,” vol. 18, no. 12, pp. 5613–5629, Dec. 2019.
• [2] E. Leitinger, S. Grebien, and K. Witrisal, “Multipath-based SLAM exploiting AoA and amplitude information,” in Proc. IEEE ICCW-19, Shanghai, China, May 2019, pp. 1–7.
• [3] J. Pinto, G. Hess, W. Ljungbergh, Y. Xia, H. Wymeersch, and L. Svensson, “Can deep learning be applied to model-based multi-object tracking?” ArXiv e-prints, 2022.

Your Tasks:

• Design and implement a deep learning algorithm based on Transformer architectures for RF-SLAM in line with the Transformer-based multi-object tracking architecture in [3].
• Evaluate the performance of the developed method and compare the performance against existing model-based SOTA algorithms [1, 2].

What we expect from you:

• You are master student in information and computer engineering, electrical engineering, audio engineering or similar studies.
• Basic knowledge in statistical signal processing, graphical models, and Bayesian modelling are beneficial.
• Programming knowledge in Matlab and Python

Contact

If you are interested and want to know more about it, send me an email to erik.leitinger@tugraz.at.