A Smartphone Multisensory Dataset for Benchmarking Urban Pedestrian Positioning Algorithms
PedROAM (Pedestrian Raw-GNSS Observations and Android Multisensor Measurements) provides raw GNSS and Android sensor measurements from natural handheld pedestrian walks in Hong Kong, with independent per-participant reference trajectories. Seven sessions collected on 21–23 April 2026 cover light, middle, deep and harsh urban environments, with 13 Google Pixel 10 Pro streams.
The dataset provides 8.90 h across 13 recordings, with individual durations of 24.0–64.1 min and assessment routes totaling 16.67 km. The uploaded directory inventory contains 9.85 GB of stored files, including compressed per-session 3D model ZIPs.
- Sensor setup
- Hong Kong sessions
- Session details and routes
- Browse and download
- GT precision reference: IE standard deviations
- Getting started
- Citation and terms
Participants walked naturally with phones held approximately horizontally at chest level, broadly aligned with walking direction. The independent reference cart followed participants wearing identity-coded ArUco boards on their backs.
| Measurement | Device / system | Available data |
|---|---|---|
| Phone GNSS, nominal 1 Hz | Google Pixel 10 Pro, Android 16; GNSSLogger | Converted RINEX 3.05, native RINEX 4.01, NMEA; Android raw logs in 12/13 streams |
| Motion, nominal 100 Hz | Phone accelerometer, gyroscope, magnetometer and orientation | Original SensorLogger CSVs, calibrated/uncalibrated streams where recorded |
| Other phone sensors | Gravity, compass, barometer, system location and pedometer | Native sensor CSVs and original recording metadata |
| Pedestrian reference | NovAtel SPAN CPT7 GNSS/INS + GeoScan 2 camera/LiDAR + back-mounted boards | Existing matched per-participant WGS-84 trajectory CSVs |
| GT precision reference | Inertial Explorer (IE) GNSS/INS solution | Per-session position and attitude standard-deviation exports in recordings/IE-STDs/ |
| Environment | Hong Kong Lands Department | Assessment-track sky masks, per-session 3D models and shared official GeoJSON pedestrian network |
The reference point is each pedestrian's back-mounted fiducial-board center, with no fixed phone-to-body lever arm. Raw rig images, LiDAR recordings, mixed rosbags and algorithm intermediates are excluded.
Browse recordings to inspect phones/, reference/ and navigation/. Environment folders contain skymask.npz and 3d.zip. Sizes below are stored file totals, including the compressed models; Dropbox-generated folder ZIP sizes may differ.
| Session | Urban setting | Phones | Recording duration / route length | Recordings | Environment |
|---|---|---|---|---|---|
| 0421-AM-2026 | Deep | 2 | P01: 61.7 min P02: 61.8 min Route: 3.50 km |
Browse · Download (949.0 MB) | Browse · Download (579.4 MB) · 3D ZIP · Sky mask |
| 0421-PM-2026 | Deep | 2 | P02: 60.8 min P03: 61.0 min Route: 2.94 km |
Browse · Download (946.3 MB) | Browse · Download (580.0 MB) · 3D ZIP · Sky mask |
| 0422-AM1-2026 | Middle | 2 | P02: 35.6 min P04: 35.5 min Route: 2.08 km |
Browse · Download (515.8 MB) | Browse · Download (607.8 MB) · 3D ZIP · Sky mask |
| 0422-AM2-2026 | Light | 2 | P02: 24.2 min P04: 24.0 min Route: 1.31 km |
Browse · Download (395.6 MB) | Browse · Download (602.8 MB) · 3D ZIP · Sky mask |
| 0422-PM-2026 | Harsh | 1 | P02: 64.1 min Route: 3.61 km |
Browse · Download (490.7 MB) | Browse · Download (1.42 GB) · 3D ZIP · Sky mask |
| 0423-AM-2026 | Light | 2 | P02: 27.0 min P04: 27.0 min Route: 1.58 km |
Browse · Download (419.1 MB) | Browse · Download (728.1 MB) · 3D ZIP · Sky mask |
| 0423-PM-2026 | Harsh | 2 | P02: 25.6 min P04: 25.6 min Route: 1.65 km |
Browse · Download (382.8 MB) | Browse · Download (605.6 MB) · 3D ZIP · Sky mask |
Recording durations are listed separately for each participant. Each duration runs from the first to the last valid matched-GT timestamp, retaining internal gaps; the 8.90 h total sums all 13 recording spans.
0423-PM-2026 is the uploaded name of the selected source session 0423-PM1-2026; the separate 0423-PM2-2026 session is excluded. 0422-PM-2026 has one phone stream. The .txt raw log is missing for 0422-AM1-2026_P04, while its RINEX/NMEA remain available.
Browse all data · Download all data
Data are organized as browsable Dropbox directories with original filenames:
PedROAM/
recordings/<session>/
phones/<participant>/gnsslogger/
phones/<participant>/sensorlogger/
reference/<participant>/trajectory-gt.csv
navigation/<date-matched broadcast ephemeris>
recordings/IE-STDs/<session>-std.txt
environment/<session>/skymask.npz
environment/<session>/3d.zip # extract to 3d/<model tiles>
environment/sidewalk/<official GeoJSON files>
For one session, download its recordings folder. Its environment folder and the shared pedestrian network are optional for sensor-only studies.
The uploaded tree contains 282 files, including seven IE standard-deviation exports, with each session’s model files packed in 3d.zip, and is within Dropbox’s folder-download limits. Whole-dataset and session downloads are both available. Extract each model ZIP beside its skymask.npz to recover 3d/.
Seven Inertial Explorer (IE) standard-deviation (STD) files, one per session, are provided in recordings/IE-STDs/ (2.14 MB total) for users assessing GT precision. Browse recordings, open IE-STDs, and download the folder or individual <session>-std.txt files. These files are also included in Download all data; they are stored separately from the individual session folders.
The exports contain GPS week (Week), GPS seconds of week (GPSTime), date (Date, month/day/year), position standard deviations (SDEast, SDNorth, SDHeight, in metres), and attitude standard deviations (RollSD, PitchSD, HdngSD, in degrees). Match the session and GPS timestamps when comparing them with reference trajectories. The reported standard deviations describe the IE GNSS/INS solution and provide a reference for GT precision; they do not include the full uncertainty of the derived pedestrian board-center trajectories.
The route previews below show the selected assessment tracks in the supplied Hong Kong 3D environment. Click a route image to view it at full resolution.
Building coverage is evaluated within 100 m of assessment positions; the tall-model fraction counts objects with vertical extent above 50 m. Mean mask elevation angle (MMEA) describes modeled sky obstruction. These indicators describe this dataset's urban categories, not universal thresholds. Sky masks are not per-phone satellite LOS/NLOS labels. Model imagery: Hong Kong Lands Department.
|
Deep urban setting Building coverage: 47.83% Recordings Environment |
|
Deep urban setting Building coverage: 38.68% Recordings Environment |
|
Middle urban setting Building coverage: 43.99% Recordings Environment |
|
Light urban setting Building coverage: 19.95% Recordings Environment |
|
Harsh urban setting Building coverage: 43.38% Recordings Environment |
|
Light urban setting Building coverage: 15.90% Recordings Environment |
|
Harsh urban setting Building coverage: 42.81% Recordings Environment |
Retain native phone timestamps and use the canonical RINEX epochs associated with the matched reference. Historical _v302.26o filenames declare RINEX 3.05 in their headers; native RINEX files declare 4.01. They represent the same stream.
The supplied code is the sky-mask decoder. Use read_skymask_frame(path, frame_index) to read a frame with its timestamp, row identifier and validity. Other data are provided in their native formats.
The paper is in preparation: PedROAM: A Smartphone Multisensory Dataset for Benchmarking Urban Pedestrian Positioning Algorithms. Final author/citation metadata and project code/data licenses are pending.
Third-party environment data retain their original provenance and applicable terms. The eight shared pedestrian-network GeoJSON files are official Hong Kong Lands Department 3D Pedestrian Network files downloaded from the CSDI portal, March 2026 version. PedROAM provides them unchanged, with their downloaded names and content.
For questions, open a GitHub issue with the session and stream ID.