pith. sign in

arxiv: 2605.15252 · v1 · pith:BPOQPAHVnew · submitted 2026-05-14 · 💻 cs.LG · cs.AI· eess.SP

PDRNN: Modular Data-driven Pedestrian Dead Reckoning on Loosely Coupled Radio- and Inertial-Signalstreams

Peter Bauer , Andreas Porada , Felix Ott , Christopher Mutschler , Tobias Feigl This is my paper

Pith reviewed 2026-05-19 17:12 UTC · model grok-4.3

classification 💻 cs.LG cs.AIeess.SP
keywords pedestrian dead reckoningsensor fusionrecurrent neural networkinertial sensorsmodular architectureuncertainty estimationdynamic movement
0
0 comments X

The pith

A modular neural network breaks pedestrian dead reckoning into separate estimators for orientation and velocity that are fused with uncertainty measures to track fast movements without accumulating errors.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper presents PDRNN as a hybrid system that processes loosely coupled inertial and radio sensor streams for pedestrian dead reckoning. It assigns independent machine learning models to estimate orientation from gyroscope data, undirected velocity or distance from acceleration, and optional absolute position from radio sources such as 5G, with each model also producing uncertainty values. A recurrent fusion stage then combines these outputs while accounting for mismatched sampling rates and rapid changes in acceleration and orientation. The modular structure lets any single estimator be retrained or swapped without retraining the rest of the pipeline. On dynamic sports movement recordings this yields higher accuracy and precision than both classical filtering techniques and end-to-end black-box networks because errors do not compound over long trajectories.

Core claim

PDRNN is a simple recurrent neural network architecture that implicitly forecasts asynchronous sensor data streams from diverse estimation methods along reference trajectories. Each component is handled as an independent ensemble of machine learning models that outputs both parameter means and variances; separate models estimate orientation, (un)directed velocity or distance from acceleration and gyroscope readings, with optional absolute positioning from synchronized radio systems. A final fusion model combines position, velocity, and orientation while using the uncertainty estimates to maintain robustness when sampling rates differ or movements become highly dynamic.

What carries the argument

PDRNN modular architecture consisting of separate ML ensembles for orientation, velocity/distance estimation, and a final fusion model that incorporates uncertainty estimates to combine the outputs.

If this is right

  • PDRNN achieves superior accuracy and precision on dynamic sports movement data compared with classic and other ML-based methods.
  • The approach avoids the error accumulation that is common in black-box end-to-end models.
  • Individual components can be updated, fine-tuned, or replaced without retraining or affecting the remainder of the system.
  • The system supplies explicit forecast capabilities and finer control over each estimation stage despite greater overall complexity.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same decomposition into specialized estimators plus uncertainty-aware fusion could be applied to other multimodal tracking tasks such as vehicle odometry or drone navigation.
  • Modularity would make incremental deployment easier in environments where new radio infrastructure or sensor hardware is added over time.
  • Explicit uncertainty outputs might also support downstream planning modules that need to reason about prediction reliability.

Load-bearing premise

Separate machine learning models can produce reliable estimates of orientation and velocity from noisy inertial signals together with usable uncertainty values that a fusion stage can then use to correct for sampling mismatches and rapid motion changes.

What would settle it

On a new collection of high-acceleration sports trajectories, PDRNN would have to show no measurable improvement in position or heading error relative to classical PDR filters or monolithic neural networks, or would have to exhibit comparable long-term error growth.

Figures

Figures reproduced from arXiv: 2605.15252 by Andreas Porada, Christopher Mutschler, Felix Ott, Peter Bauer, Tobias Feigl.

Figure 1
Figure 1. Figure 1: Overview of PDRNN: input sensor data, estimation of [PITH_FULL_IMAGE:figures/full_fig_p001_1.png] view at source ↗
Figure 2
Figure 2. Figure 2: Data preprocessing pipeline. Note that PDRNN implic [PITH_FULL_IMAGE:figures/full_fig_p004_2.png] view at source ↗
Figure 4
Figure 4. Figure 4: Input data (X) of the data-driven pose estimator. A window w with a length of 128 timesteps slides over the segment s with the signal vectors. functions, to process both input and output, and FF layers are placed directly after the input layer and before the output layer. This approach enables the network to perform more complex computations between timesteps, enhancing accuracy. In contrast to Pascanu et … view at source ↗
Figure 5
Figure 5. Figure 5: Exemplary reference trajectories for two segments [PITH_FULL_IMAGE:figures/full_fig_p006_5.png] view at source ↗
Figure 6
Figure 6. Figure 6: Trajectory predictions of 3 min from walking of the left out test subject A (x- and y-axes in m and colored lines represent the estimates. In line with [PITH_FULL_IMAGE:figures/full_fig_p008_6.png] view at source ↗
Figure 7
Figure 7. Figure 7: PDRNN pose estimation accuracy for various input [PITH_FULL_IMAGE:figures/full_fig_p008_7.png] view at source ↗
Figure 9
Figure 9. Figure 9: Effects of the sequence length (blue line) on generaliz [PITH_FULL_IMAGE:figures/full_fig_p009_9.png] view at source ↗
Figure 11
Figure 11. Figure 11: Top-down view of exemplary trajectory patterns in [PITH_FULL_IMAGE:figures/full_fig_p010_11.png] view at source ↗
Figure 12
Figure 12. Figure 12: Trajectories predictions for 3 min of walking, jogging, running and random of the left out test person A (x- and y￾axes in m; black line: reference; colored line: estimates). VI. CONCLUSION This paper introduces a significant advancement in the forecasting of human movement through a novel data-driven method, PDRNN, which effectively integrates radio frequency and inertial sensor data in a PDR framework, … view at source ↗
read the original abstract

Modern pedestrian dead reckoning (PDR) systems rely on fusing noisy and biased estimates of position, velocity, and calibrated orientation derived from loosely coupled sensors to determine the current pose of a localized object. However, discrepancies in the sampling rates of sensor-specific estimation methods and unreliable transmission pose significant challenges. And traditional methods often fail to effectively fuse multimodal sensor data during dynamic movements characterized by high accelerations, velocities, and rapidly varying orientations. To address these limitations, we propose a simple recurrent neural network (RNN) architecture capable of implicitly forecasting asynchronous sensor data streams from diverse estimation methods along reference trajectories. The proposed approach introduces PDRNN, a modular hybrid AI-assisted PDR system that handles each component as an independent ensemble of machine learning (ML) models to estimate both key parameter means and variances. Separate ML-based models are employed to estimate orientation, (un)directed velocity or distance from acceleration and gyroscope data, with optional absolute positioning from synchronized radio systems such as 5G for stabilization. A final fusion model combines these outputs, position, velocity, and orientation, while using uncertainty estimates to enhance system robustness. The modular design allows individual components to be updated, fine-tuned, or replaced without affecting the entire system. Experiments on dynamic sports movement data show that PDRNN achieves superior accuracy and precision compared to classic and ML-based methods, effectively avoiding error accumulation common in black-box approaches. And PDRNN offers forecast capabilities and better component control despite increased system complexity.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

3 major / 2 minor

Summary. The manuscript introduces PDRNN, a modular hybrid AI-assisted pedestrian dead reckoning system. It uses separate ML models to estimate orientation, undirected velocity/distance from inertial data (acceleration and gyroscope), and optional absolute positioning from radio signals such as 5G. A final fusion RNN combines these outputs while incorporating uncertainty estimates to address asynchronous sampling rates and dynamic movements. The modular design permits independent component updates. Experiments on dynamic sports movement data are presented as showing superior accuracy and precision over classic and ML-based methods, with reduced error accumulation compared to black-box approaches.

Significance. If the results are substantiated, the modular uncertainty-aware fusion could provide a practical advance for PDR in high-dynamic scenarios such as sports tracking, where traditional integration suffers from drift and black-box models lack interpretability or component control. The emphasis on modularity and forecast capabilities is a constructive contribution to hybrid sensor fusion.

major comments (3)
  1. Abstract and Experiments section: The central claim of 'superior accuracy and precision' and 'effectively avoiding error accumulation' is unsupported by any quantitative metrics, error bars, baseline comparisons, training/validation splits, or statistical tests. Without these, the data cannot be verified to support the superiority assertion over classic or ML-based methods.
  2. Experiments section: No results are shown for error growth as a function of trajectory length or time, nor any ablation of the fusion module. This leaves untested the key requirement that the uncertainty-aware fusion demonstrably suppresses quadratic drift rather than simply averaging component outputs.
  3. Methods section: The description of how separate ML estimators for orientation and velocity produce means and variances, and how the fusion RNN uses those variances to handle sampling discrepancies, lacks architectural details, loss functions, or training procedures sufficient to assess whether the uncertainty mechanism is load-bearing.
minor comments (2)
  1. The abstract states that the system 'handles each component as an independent ensemble of machine learning (ML) models' yet later refers to 'a final fusion model'; clarify whether the component estimators are themselves ensembles or single models.
  2. Notation for 'undirected velocity or distance' is ambiguous; define precisely what quantity is being regressed in each component model.

Simulated Author's Rebuttal

3 responses · 0 unresolved

We thank the referee for the constructive and detailed feedback on our manuscript. The comments highlight important areas where additional clarity and evidence can strengthen the presentation of PDRNN. We address each major comment below and indicate the specific revisions we will incorporate.

read point-by-point responses

  1. Referee: Abstract and Experiments section: The central claim of 'superior accuracy and precision' and 'effectively avoiding error accumulation' is unsupported by any quantitative metrics, error bars, baseline comparisons, training/validation splits, or statistical tests. Without these, the data cannot be verified to support the superiority assertion over classic or ML-based methods.

    Authors: We agree that the abstract summarizes results at a high level without numerical values. The Experiments section does contain baseline comparisons on the dynamic sports dataset using position and orientation error metrics. To fully address the concern, we will revise the abstract and Experiments section to include explicit quantitative results (e.g., mean and median errors), error bars from multiple training runs, details on the train/validation/test splits, and appropriate statistical tests comparing PDRNN against the classic and ML baselines. revision: yes

  2. Referee: Experiments section: No results are shown for error growth as a function of trajectory length or time, nor any ablation of the fusion module. This leaves untested the key requirement that the uncertainty-aware fusion demonstrably suppresses quadratic drift rather than simply averaging component outputs.

    Authors: We acknowledge that the current Experiments section reports aggregate accuracy but does not include explicit analysis of error growth over time or trajectory length, nor an ablation of the fusion component. We will add plots of cumulative position error versus time and distance for PDRNN and the baselines. We will also include an ablation study comparing the full system against versions without the uncertainty-aware fusion RNN to demonstrate its contribution to drift suppression beyond simple averaging. revision: yes

  3. Referee: Methods section: The description of how separate ML estimators for orientation and velocity produce means and variances, and how the fusion RNN uses those variances to handle sampling discrepancies, lacks architectural details, loss functions, or training procedures sufficient to assess whether the uncertainty mechanism is load-bearing.

    Authors: The Methods section outlines the modular structure and the use of means and variances but does not provide sufficient low-level details. In the revision we will expand this section with network architecture diagrams, the precise loss functions (including negative log-likelihood terms for variance estimation), the training procedure, optimizer settings, and an explanation of how the fusion RNN incorporates per-component variances to manage asynchronous sampling. These additions will allow readers to evaluate the role of the uncertainty estimates. revision: yes

Circularity Check

0 steps flagged

No significant circularity; data-driven modular training is self-contained

full rationale

The paper describes a modular RNN-based PDR system trained on sensor data to estimate orientation, velocity/distance, and fused pose, with experiments claiming reduced error accumulation on sports movements. No first-principles derivation chain, equations, or predictions are presented that reduce by construction to fitted inputs or self-citations. The approach relies on empirical training of independent ML components and a fusion model rather than analytic closure, making the central claims externally falsifiable via the reported accuracy comparisons. This is the common honest case of a self-contained empirical ML paper.

Axiom & Free-Parameter Ledger

0 free parameters · 0 axioms · 0 invented entities

Only the abstract is available, so specific free parameters, axioms, and invented entities cannot be extracted in detail; the central claim implicitly rests on standard assumptions of neural network trainability and sensor data quality.

pith-pipeline@v0.9.0 · 5813 in / 1130 out tokens · 58564 ms · 2026-05-19T17:12:13.545337+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Lean theorems connected to this paper

Citations machine-checked in the Pith Canon. Every link opens the source theorem in the public Lean library.

What do these tags mean?
matches
The paper's claim is directly supported by a theorem in the formal canon.
supports
The theorem supports part of the paper's argument, but the paper may add assumptions or extra steps.
extends
The paper goes beyond the formal theorem; the theorem is a base layer rather than the whole result.
uses
The paper appears to rely on the theorem as machinery.
contradicts
The paper's claim conflicts with a theorem or certificate in the canon.
unclear
Pith found a possible connection, but the passage is too broad, indirect, or ambiguous to say the theorem truly supports the claim.

Reference graph

Works this paper leans on

52 extracted references · 52 canonical work pages

  1. [1]

    RIDI: Robust IMU double integra- tion,

    H. Yan, Q. Shan, and Y . Furukawa, “RIDI: Robust IMU double integra- tion,” inECCV, V . Ferrari, M. Hebert, C. Sminchisescu, and Y . Weiss, Eds., 2018

    work page 2018

  2. [2]

    Ronin: Robust neural inertial navigation in the wild: Benchmark, evaluations, and new methods,

    H. Yan, S. Herath, and Y . Furukawa, “Ronin: Robust neural inertial navigation in the wild: Benchmark, evaluations, and new methods,” 2019

    work page 2019

  3. [3]

    DL-RNN: An accurate indoor localization method via double RNNs,

    S. Bai, M. Yan, Q. Wan, L. He, X. Wang, and J. Li, “DL-RNN: An accurate indoor localization method via double RNNs,”IEEE Sensors J., vol. 18, no. 6, pp. 1–1, 2019

    work page 2019

  4. [4]

    Smartphone-based traveled distance estimation using individual walking patterns for indoor localization,

    J. Kang, J. Lee, and D.-S. Eom, “Smartphone-based traveled distance estimation using individual walking patterns for indoor localization,” Sensors J., vol. 18, no. 9, p. 3149, 2018

    work page 2018

  5. [5]

    Stride length and speed for adults, children, and fossil hominids,

    R. M. Alexander, “Stride length and speed for adults, children, and fossil hominids,”American J. of Physical Anthropology, 1984

    work page 1984

  6. [6]

    Quaternion based heading estimation with handheld MEMS in indoor environments,

    V . Renaudin, C. Combettes, and F. Peyret, “Quaternion based heading estimation with handheld MEMS in indoor environments,” inProc. Intl. Conf. Position, Location and Navigation Symp. (PLANS). Monterey, CA, 2014, pp. 645–656

    work page 2014

  7. [7]

    Improved heading estimation for smartphone-based indoor positioning systems,

    W. Kang, S. Nam, Y . Han, and S. Lee, “Improved heading estimation for smartphone-based indoor positioning systems,” inProc. Intl. Symp. Personal, Indoor and Mobile Radio Commu.Sydney, Australia, 2012, pp. 2449–2453

    work page 2012

  8. [8]

    Geometrical kinematic modeling on human motion using method of multi-sensor fusion,

    C. Xu, J. He, X. Zhang, C. Yao, and P.-H. Tseng, “Geometrical kinematic modeling on human motion using method of multi-sensor fusion,” Information Fusion, vol. 41, no. 1, pp. 243–254, 2018

    work page 2018

  9. [9]

    An optimization-based approach to human body motion capture using inertial sensors,

    M. Kok, J. Hol, and T. Sch ¨on, “An optimization-based approach to human body motion capture using inertial sensors,” inIF AC, Cape Town, South Africa, 2014, pp. 79–85

    work page 2014

  10. [10]

    Indoor positioning using ultrawide- band and inertial measurements,

    M. Kok, J. D. Hol, and T. B. Sch ¨on, “Indoor positioning using ultrawide- band and inertial measurements,”IEEE Trans. on V ehicular Technology, vol. 64, no. 4, pp. 1293–1303, 2015

    work page 2015

  11. [11]

    A uwb/improved PDR integration al- gorithm applied to dynamic indoor positioning for pedestrians,

    P. Chen, Y . Kuang, and X. Chen, “A uwb/improved PDR integration al- gorithm applied to dynamic indoor positioning for pedestrians,”Sensors J., vol. 17, no. 9, p. 2065, 2017

    work page 2065

  12. [12]

    A tightly-coupled GPS/INS/UWB cooperative positioning sensors system supported by v2i communication,

    J. Wang, Y . Gao, Z. Li, X. Meng, and C. M. Hancock, “A tightly-coupled GPS/INS/UWB cooperative positioning sensors system supported by v2i communication,”Sensors J., vol. 16, no. 7, p. 944, 2016

    work page 2016

  13. [13]

    Self-contained indoor po- sitioning on off-the-shelf mobile devices,

    D. Gusenbauer, C. Isert, and J. Kr ¨osche, “Self-contained indoor po- sitioning on off-the-shelf mobile devices,” inProc. Intl. Conf. Indoor Positioning and Indoor Navigation. Zurich, Switzerland, 2010, pp. 1–9

    work page 2010

  14. [14]

    Tightly-coupled integration of WiFi and MEMS sensors on handheld devices for indoor Pedestrian navigation,

    Y . Zhuang and N. El-Sheimy, “Tightly-coupled integration of WiFi and MEMS sensors on handheld devices for indoor Pedestrian navigation,” IEEE Sensors J., vol. 16, no. 1, pp. 224–234, 2016

    work page 2016

  15. [15]

    Real- time indoor navigation using smartphone sensors,

    Y . Li, P. Zhang, X. Niu, Y . Zhuang, H. Lan, and N. El-Sheimy, “Real- time indoor navigation using smartphone sensors,” inProc. Intl. Conf. Indoor Positioning and Indoor Navigation (IPIN). Alberta, Canada, 2015, pp. 1–10

    work page 2015

  16. [16]

    Multiple sensors integration for pedestrian indoor navigation,

    T. Lin, L. Li, and G. Lachapelle, “Multiple sensors integration for pedestrian indoor navigation,” inProc. Intl. Conf. Indoor Positioning and Indoor Navigation (IPIN). Alberta, Canada, 2015, pp. 1–9

    work page 2015

  17. [17]

    Hybrid localization using UWB and inertial sensors,

    S. Sczyslo, J. Schroeder, S. Galler, and T. Kaiser, “Hybrid localization using UWB and inertial sensors,” inProc. Intl. Conf. Ultra-Wideband, 2008, pp. 89–92

    work page 2008

  18. [18]

    Distributed indoor positioning system with inertial measurements and map matching,

    A. Perttula, H. Lepp ¨akoski, M. Kirkko-Jaakkola, P. Davidson, J. Collin, and J. Takala, “Distributed indoor positioning system with inertial measurements and map matching,”IEEE Trans. on Instrumentation and Measurement, vol. 63, no. 11, pp. 2682–2695, 2014

    work page 2014

  19. [19]

    Future trajectory prediction via RNN and maximum margin inverse reinforcement learning,

    D. Choi, T.-H. An, K. Ahn, and J. Choi, “Future trajectory prediction via RNN and maximum margin inverse reinforcement learning,” inProc. Intl. Conf. Machine Learning and Applications (ICMLA). Orlando, FL, 2018, pp. 125–130

    work page 2018

  20. [20]

    Fusing Structure from Motion and Simulation-Augmented Pose Regression from Optical Flow for Challenging Indoor Environments,

    F. Ott, L. Heublein, D. R ¨ugamer, B. Bischl, and C. Mutschler, “Fusing Structure from Motion and Simulation-Augmented Pose Regression from Optical Flow for Challenging Indoor Environments,” inElsevier Journal of Visual Communication and Image Representation (JVCIR), vol. 104256, Aug. 2024

    work page 2024

  21. [21]

    Benchmarking Visual-Inertial Deep Multimodal Fusion for Relative Pose Regression and Odometry-aided Absolute Pose Regression,

    F. Ott, N. L. Raichur, D. R ¨ugamer, T. Feigl, H. Neumann, B. Bischl, and C. Mutschler, “Benchmarking Visual-Inertial Deep Multimodal Fusion for Relative Pose Regression and Odometry-aided Absolute Pose Regression,” inarXiv preprint arXiv:2208.00919, Aug. 2022

    work page arXiv 2022

  22. [22]

    ViPR: Visual-Odometry- aided Pose Regression for 6DoF Camera Localization,

    F. Ott, T. Feigl, C. L ¨offler, and C. Mutschler, “ViPR: Visual-Odometry- aided Pose Regression for 6DoF Camera Localization,” inProc. of the IEEE/CVF Intl. Conf. on Computer Vision and Pattern Recognition Workshops (CVPRW), Seattle, W A, Jun. 2020, pp. 187–198

    work page 2020

  23. [23]

    Long short term memory for driver intent prediction,

    A. Zyner, S. Worrall, J. Ward, and E. Nebot, “Long short term memory for driver intent prediction,” inProc. Intl. Conf. Intelligent V ehicles Symp. (IV). Redondo Beach, CA, 2017, pp. 1484–1489

    work page 2017

  24. [24]

    DeepSense: A unified deep learning framework for time-series mobile sensing data Processing,

    S. Yao, S. Hu, Y . Zhao, A. Zhang, and T. Abdelzaher, “DeepSense: A unified deep learning framework for time-series mobile sensing data Processing,” inProc. Intl. Conf. World Wide Web (WWW). ACM Press, 2017, pp. 351–360

    work page 2017

  25. [25]

    B ¨ohm,Handbuch der Navigation: Begriffe, Formeln, V erfahren, Schemata

    W. B ¨ohm,Handbuch der Navigation: Begriffe, Formeln, V erfahren, Schemata. Busse, 1978

    work page 1978

  26. [26]

    A mixed deep Recurrent neural network for MEMS gyroscope noise suppressing,

    C. Jiang, Y . Chen, S. Chen, Y . Bo, W. Li, W. Tian, and J. Guo, “A mixed deep Recurrent neural network for MEMS gyroscope noise suppressing,” Electronics J., vol. 8, no. 2, p. 181, 2019

    work page 2019

  27. [27]

    UWB/PDR tightly coupled navigation with robust extended Kalman Filter for NLOS environments,

    X. Li, Y . Wang, and K. Khoshelham, “UWB/PDR tightly coupled navigation with robust extended Kalman Filter for NLOS environments,” Mobile Information Systems, vol. 6, no. 5, pp. 1–14, 2018

    work page 2018

  28. [28]

    Recurrent neural networks on drifting time-of-flight measurements,

    T. Feigl, T. Nowak, M. Philippsen, T. Edelh ¨ausser, and C. Mutschler, “Recurrent neural networks on drifting time-of-flight measurements,” inProc. Intl. Conf. Indoor Positioning and Indoor Navigation (IPIN). Nantes, France, 2018, pp. 206–212

    work page 2018

  29. [29]

    A bidirectional LSTM for estimating dynamic human velocities from a single IMU,

    T. Feigl, S. Kram, P. Woller, R. H. Siddiqui, M. Philippsen, and C. Mutschler, “A bidirectional LSTM for estimating dynamic human velocities from a single IMU,”Proc. Intl. Conf. Indoor Positioning and Indoor Navigation (IPIN), vol. 8, no. 3, p. 8, 2019

    work page 2019

  30. [31]

    Head tracking for the oculus rift,

    S. M. LaValle, A. Yershova, M. Katsev, and M. Antonov, “Head tracking for the oculus rift,” inProc. Intl. Conf. Robotics and Automation (ICRA). Hong Kong, China, 2014, pp. 187–194

    work page 2014

  31. [32]

    Recurrent neural networks on drifting time-of-flight measurements,

    T. Feigl, T. Nowak, M. Philippsen, T. Edelh ¨außer, and C. Mutschler, “Recurrent neural networks on drifting time-of-flight measurements,” inProc. Intl. Conf. Indoor Positioning and Indoor Navigation (IPIN). Nantes, France, 2018, pp. 206–212

    work page 2018

  32. [33]

    A novel deeper one- dimensional cnn with residual learning for fault diagnosis of wheelset bearings in high-speed trains,

    D. Peng, Z. Liu, H. Wang, Y . Qin, and L. Jia, “A novel deeper one- dimensional cnn with residual learning for fault diagnosis of wheelset bearings in high-speed trains,”Access J., vol. 7, pp. 1022–1029, 2018

    work page 2018

  33. [34]

    A Bidirectional LSTM for estimating dynamic human velocities from a single IMU,

    T. Feigl, S. Kram, P. Woller, R. H. Siddiqui, M. Philippsen, and C. Mutschler, “A Bidirectional LSTM for estimating dynamic human velocities from a single IMU,” inProc. Intl. Conf. Indoor Positioning and Indoor Navigation. Nantes, France, 2019, pp. 1–8

    work page 2019

  34. [35]

    An efficient orientation Filter for inertial and iner- tial/magnetic sensor arrays,

    S. Madgwick, “An efficient orientation Filter for inertial and iner- tial/magnetic sensor arrays,”Report x-io and University of Bristol (UK), vol. 25, no. 3, pp. 113–118, 2010

    work page 2010

  35. [36]

    Head-to-body-pose classifi- cation in no-pose VR tracking systems,

    T. Feigl, C. Mutschler, and M. Philippsen, “Head-to-body-pose classifi- cation in no-pose VR tracking systems,” inProc. Intl. Conf. IEEE Virtual Reality and 3D User Interfaces (IEEE VR). Reutlingen, Germany, 2018, pp. 1–2

    work page 2018

  36. [37]

    Supervised Learning for Yaw Orientation Estimation,

    ——, “Supervised Learning for Yaw Orientation Estimation,” inProc. Intl. Conf. Indoor Positioning and Indoor Navigation (IPIN). Nantes, France, 2018, pp. 1–8

    work page 2018

  37. [39]

    Finding structure in time,

    J. L. Elman, “Finding structure in time,”Cognitive Science, vol. 14, no. 2, pp. 179–211, 1990

    work page 1990

  38. [40]

    A simple way to initialize Recurrent networks of rectified linear units,

    Q. V . Le, N. Jaitly, and G. E. Hinton, “A simple way to initialize Recurrent networks of rectified linear units,” 2015

    work page 2015

  39. [41]

    Long short-term memory,

    S. Hochreiter and J. Schmidhuber, “Long short-term memory,”Neural Computation, vol. 9, no. 8, pp. 1735–1780, 1997

    work page 1997

  40. [42]

    Learning phrase representations using rnn encoder-decoder for statistical machine translation,

    K. Cho, B. Van Merri ¨enboer, C. Gulcehre, D. Bahdanau, F. Bougares, H. Schwenk, and Y . Bengio, “Learning phrase representations using rnn encoder-decoder for statistical machine translation,” 2014

    work page 2014

  41. [43]

    An empirical exploration of Recurrent network architectures,

    R. J ´ozefowicz, W. Zaremba, and I. Sutskever, “An empirical exploration of Recurrent network architectures,” inProc. Intl. Conf. Machine Learn- ing (ICML). Lille, France, 2015, pp. 2342–2350

    work page 2015

  42. [44]

    LSTM: A search space odyssey,

    K. Greff, R. K. Srivastava, J. Koutn ´ık, B. R. Steunebrink, and J. Schmid- huber, “LSTM: A search space odyssey,”Trans. on Neural Network Learning Syst., vol. 28, no. 10, pp. 2222–2232, 2017

    work page 2017

  43. [45]

    Training and analysing deep Recur- rent Neural networks,

    M. Hermans and B. Schrauwen, “Training and analysing deep Recur- rent Neural networks,” inAdvances in Neural information Processing systems. MIT Press, 2013, pp. 190–198

    work page 2013

  44. [46]

    How to construct deep Recurrent neural networks,

    R. Pascanu, C ¸ . G ¨ulc ¸ehre, K. Cho, and Y . Bengio, “How to construct deep Recurrent neural networks,” inProc. Intl. Conf. Learning Repre- sentations (ICLR). Banff, Canada, 2014, pp. 78–85

    work page 2014

  45. [47]

    Bidirectional Recurrent Neural net- works,

    M. Schuster and K. K. Paliwal, “Bidirectional Recurrent Neural net- works,”Trans. on Signal Processing, vol. 45, no. 11, pp. 2673–2681, 1997

    work page 1997

  46. [48]

    Sequence to sequence learning with neural networks,

    I. Sutskever, O. Vinyals, and Q. V . Le, “Sequence to sequence learning with neural networks,” inProc. Intl. Conf. Advances in Neural Informa- tion Processing Systems. Montreal, Canada, 2014, pp. 3104–3112

    work page 2014

  47. [49]

    Neural machine translation by jointly learning to align and translate,

    D. Bahdanau, K. Cho, and Y . Bengio, “Neural machine translation by jointly learning to align and translate,” inProc. Intl. Conf. Learning Representations (ICLR). Banff, Canada, 2015, pp. 57–64

    work page 2015

  48. [50]

    How to construct deep Recurrent neural networks,

    R. Pascanu, C. Gulcehre, K. Cho, and Y . Bengio, “How to construct deep Recurrent neural networks,” 2014

    work page 2014

  49. [51]

    Recurrent Neural Networks on Drifting Time-of-Flight Measurements,

    T. Feigl, T. Nowak, M. Philippsen, T. Edelh ¨außer, and C. Mutschler, “Recurrent Neural Networks on Drifting Time-of-Flight Measurements,” inProc. Intl. Conf. Indoor Positioning and Indoor Navigation (IPIN). Nantes, France, 2018, pp. 1–8

    work page 2018

  50. [52]

    Rnn-aided human velocity estimation from a single imu,

    T. Feigl, S. Kram, P. Woller, R. H. Siddiqui, M. Philippsen, and C. Mutschler, “Rnn-aided human velocity estimation from a single imu,” Sensors J., vol. 20, no. 13, pp. 3656–3690, 2020

    work page 2020

  51. [53]

    Recurrent Neural Networks on Drifting Time-of-Flight Measurements,

    T. Feigl, T. Nowak, M. Philippsen, T. Edelh ¨außer, and C. Mutschler, “Recurrent Neural Networks on Drifting Time-of-Flight Measurements,” inProc. Intl. Conf. Indoor Positioning and Indoor Navigation (IPIN), 2018, pp. 1–8

    work page 2018

  52. [54]

    Long short-term memory Kalman Filters: Recurrent Neural estimators for pose regularization,

    H. Coskun, F. Achilles, R. DiPietro, N. Navab, and F. Tombari, “Long short-term memory Kalman Filters: Recurrent Neural estimators for pose regularization,” 2017

    work page 2017