Highly accurate, robust, and continuous localization is essential for the safe navigation and control of autonomous vehicles. Modern autonomous systems, including ROS 2-based platforms, rely on heterogeneous localization methods operating at different sampling rates and with different timing characteristics. Ignoring their temporal differences can lead to misleading trajectory error statistics. This paper presents an ROS 2-based localization evaluation framework that applies the spatio-temporal alignment of Tombrink et al. to asynchronous RTK-GNSS and LiDAR-based NDT trajectories using a sliding-window strategy. The framework was evaluated on a 767 s, approximately 3.67 km urban vehicle recording. In a same-solver comparison using samples not included in the preceding fitting windows, temporal compensation reduced horizontal ATE RMSE from 1.207 m to 0.740 m (38.7%). The results also distinguish local fitting errors from fixed-transform trajectory agreement and show that the estimated time shift depends on the calibration interval. The framework thus supports a documented comparison of asynchronously acquired localization trajectories under the evaluated conditions.

The objective of this study is to examine the Radar Cross Section (RCS) of instruments designed for Autonomous Driving Systems (ADAS) testing, with the intention of comparing the results to those of actual human subjects. The RCS values of both dummy and platform objects were documented at varying distances and positions, with the objective of ascertaining the extent to which dummies can serve as substitutes for human values in vehicle radar sensing tests. The findings, substantiated by graphical representations and statistical analyses (e.g., Pearson and Spearman correlation), reveal a moderately strong positive correlation between the RCS and human values, which is statistically significant. The outcomes of the tests demonstrate that the developed instruments can substitute for real human radar cross-section values within the range of 5-15 m. However, as the distance increases, larger deviations are observed. These discrepancies underscore the necessity for a refinement of the dummy design in future ADAS tests, ensuring that distance-sensitive tests accurately reflect real human measurements.

Validation of Advanced Driver Assistance Systems (ADAS) relies on standardized pedestrian soft targets to ensure comparable safety assessment. However, real-world pedestrians differ from standardized targets in clothing and carried objects, potentially altering radar-based sensing. This exploratory study investigates how such variations affect the sensor-reported apparent radar cross section (RCS) of a pedestrian soft target. Eight configurations were evaluated on a closed proving ground. For each configuration, one 30 s stationary measurement at approximately 5 m and three vehicle-approach runs ranging from 30 to 5 m were performed using a vehicle-mounted automotive radar. Dynamic runs were evaluated in three distance bands using run-level median apparent RCS values, differences from the reference, and run-to-run standard deviation. The backpack and notebook bag produced the largest observed positive deviations, reaching +9.17 and +7.25 dB at 20–30 m. The visibility vest showed a moderate, consistent positive shift, whereas the winter coat and jacket remained closer to the reference in most bands. The umbrella configuration exhibited the greatest run-to-run variability. Among the investigated configurations, the largest observed differences were associated with the backpack and notebook bag, both of which substantially altered the external shape of the pedestrian surrogate. Because only three dynamic repetitions, one radar sensor, one pedestrian surrogate, and one side-profile orientation were investigated, the findings should be interpreted as an exploratory, sensor-specific sensitivity analysis rather than as a definitive characterization of pedestrian RCS or ADAS performance.

This study presents a reproducible GNSS/INS-based workflow for reconstructing nominal binder-spreading footprints in road construction. Four Firebase measurement exports from 16,018 recorded observations were analyzed. Data preprocessing involved outlier detection, stagnation identification, transport and refilling route detection, and consistency checks between GNSS-derived movement direction and INS orientation. Temporal continuity was assessed using a 5 s threshold, with longer measurement gaps treated as segment boundaries and marked by inferred nozzle-off states. The resulting operating-state and navigation records were converted into CAD-compatible EOV polygons. The four archived AutoLISP outputs contain 164 reconstructed footprints. Within a 6173.52 m² reference polygon, their four-day geometric union covers 5398.88 m² (87.45%), leaving 774.64 m² (12.55%) uncovered. The gross clipped polygon contribution is 17,698.40 m², including 12,299.53 m² of multiplicity-weighted redundancy. An independent 1000-replicate Monte Carlo registration-perturbation analysis, assuming independent daily EOV shifts uniformly distributed within ±0.25 m per coordinate, yielded a mean coverage of 87.57% and 2.5–97.5% simulation quantiles of 86.60–88.51%. All headline areas were independently recomputed from the unchanged AutoLISP polygon vertices and agree with the stored area metadata. The study establishes an auditable geometric basis for reviewing spatial coverage and repeated nominal application.