A Novel Integrated IMU-UWB Framework for Walking Trajectory Estimation in Non-Line-of-Sight Scenarios Involving Turning Gait

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

This paper presents a walking trajectory estimation algorithm considering the NLoS of UWB observations and turning accumulate error.  There is still several aspects I am concerned:

1) The estimated trajectory seems to be more "smooth" compared to the real trajectory especially in the turning points, I guess this may because that the eskf method may smooth the trajectory when integrating the imu and uwb observation. Will this phenomenon become more serious over time? 

2) How does the performance of the designed turning correction module compare with that of the traditional method?

3) It would better to show the UWB NLoS areas in the trajectory to see the performance of the eskf.

4) Fig.6 is kind of blurry. It is better to improve the quality of the figure.

Author Response

The authors appreciate the constructive comments made by this reviewer. We have tried to address each of these, as summarized in the attachment, and trust these have enhanced the quality of the manuscript. Specific modifications to the manuscript are indicated, and the updated locations in the manuscript are identified.

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

The article proposes an integrated IMU-UWB framework for indoor gait trajectory estimation under NLOS and turning conditions. The proposal combines an Error-State Kalman Filter (ESKF) with a turn-phase-specific correction module. To achieve this, a wearable sensor based on the BNO055 IMU and DW1000/LD150 UWB sensor is implemented with ESP32, synchronizing the data (IMU at 100 Hz and UWB at 10 Hz) using temporal interpolation. Turns are detected with a threshold on the GyroZ axis (±40 dps), and lateral displacements are corrected for pre-turn and post-turn movements. In an experiment with 18 participants, the authors report a reduction in MAE from 22.2 cm to 7.0 cm and improvements in correlation with the reference trajectory (up to 99.9% in one case).

Among its main strengths is the well-defined problem: two critical sources of error are identified (the impact of NLOS conditions and error accumulation during turns), and an approach is proposed that addresses them jointly. The experimental design is also clear and replicable, with details on sensors, sampling rates, anchor arrangement, and test conditions. Furthermore, the comparison between UWB-only, UWB-IMU with ESKF, and the full framework demonstrates the contribution of the proposed module, showing a substantial improvement in trajectory accuracy.

Overall, the work constitutes a solid contribution to the field of indoor localization, effectively integrating sensor fusion with an event-specific correction strategy. However, some adjustments, described below, are suggested to strengthen the manuscript.

1.
Consistency of results and metrics: There is a numerical inconsistency between sections. In Results/Table 1, the mean MAE increases from 22.2 cm → 7.0 cm, while in Discussion, it states 21.8 cm → 6.9 cm and a “68.3%” improvement. Unify figures, clearly define whether “improvement” refers to relative error reduction, increased precision, or percentage reduction in MAE, and report 95% CIs and paired tests (e.g., Wilcoxon) for MAE and r (Spearman). Include error bar graphs or violin plots by participant and by condition.
2.
Statistical modeling and sample size: Although the N=18, no justification for power or intersubject variability is presented. The following are suggested: (i) a priori/post-hoc power analysis, (ii) confidence intervals by condition, (iii) subgroup analysis (e.g., sex/height/mass) for minimal generalization.
3.
Detailed ESKF and noise; key specifications are missing: Q and R matrices, Kalman gain tuning strategy for NLOS, and stability validation (observability conditions). Request an annex with F, B, H, Q, R, states, and units; describe how NLOS is detected/attenuated in R or in innovation (residual gating) and whether piecewise relinearization was performed.
4.
Gyro correction: robustness and sensitivity: The 40 dps threshold in GyroZ was "iteratively determined"; A sensitivity study (e.g., 20–80 dps sweep) and a comparison against alternative detectors (adaptive filters, gait phase HMMs, or cadence-based rules) with turn detection accuracy/recall are required. Furthermore, justify the formula for the C coefficient and show its distribution (pre/post-turn) versus turn angle and intensity.
5.
External validation and challenging scenarios: The entire experiment was conducted in a laboratory setting with healthy young individuals; generalization to older adults or real-life industrial environments is open. Include (or discuss as a quantified limitation) scenarios with continuous curves, turns >90°, stairs/uneven surfaces, magnetic interference, timestamp noise, and packet loss (UWB/IMU). Ideal: a public dataset with raw signals and turn annotations.
6.
Ground-truth and temporal traceability: The "ground truth" trajectory is defined as 2D coordinates spaced 1 cm apart, but the reference method is unclear (optical system? manual marking?). It is critical to detail the ground truth system, its error (< cm), and the temporal alignment with IMU/UWB (clock drift, jitter, NTP/PTP offset). Evaluate aliasing by IMU→UWB interpolation (10×) and present a synthetic synchronization experiment.

Author Response

The authors appreciate the constructive comments and suggestions made by the reviewer. We have tried to address each of these, as summarized below, and trust these have enhanced the quality of the manuscript. Specific modifications to the manuscript are indicated here, and the updated locations in the manuscript are identified.

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

Comments and Suggestions for Authors

This study presents a novel IMU-UWB framework with ESKF and turning correction, effectively improving trajectory estimation in NLOS and turning scenarios, shown by lower MAE (7.0 cm) and higher correlation, with clear practical value. It is well written. However, two issues should be solved before publication.

1. How does the IMU-UWB framework differ from other existing methods?
2. Are there any limitations to the method?

Author Response

The authors appreciate the constructive comments and suggestions made by the reviewer. We have tried to address each of these, as summarized below, and trust these have enhanced the quality of the manuscript. Specific modifications to the manuscript are indicated here, and the updated locations in the manuscript are identified.

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

Comments and Suggestions for Authors

I don't see any novelty in this study, perhaps I can understand the work better after you answer my following two questions

1. Why you need UWB here? Because foot mounted imu with zero velocity update can already achieve a very accurate result.
2. Why don't you just use orientation integrated from gyro to indicate turning? 

Author Response

The authors appreciate the constructive comments and suggestions made by the reviewer. We have tried to address each of these, as summarized below, and trust these have enhanced the quality of the manuscript. Specific modifications to the manuscript are indicated here, and the updated locations in the manuscript are identified.

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 4 Report

Comments and Suggestions for Authors

I mean not all the fusions require magnetometers. Do you have any plots how you can detect turns directly from gyro data? Because the turn can be easier hidden in the stride signal? 

Author Response

The authors appreciate the constructive comments made by the reviewer. We have tried to address each of these, as summarized below, and trust these have enhanced the quality of the manuscript. Specific modifications to the manuscript are also indicated. 

Please see the attachment.

 

Author Response File: Author Response.pdf