No GPS required: our app can now locate underground trains

Original Article ↗ Hacker News Discussion ↗

Overall reaction to the feature & app

  • Many commenters praise the idea as “super cool” and say Transit is one of the best transit apps they use daily.
  • Others report uninstalling after inaccurate behavior on specific systems (e.g., London Underground, NYC Subway), showing mixed real‑world outcomes.
  • Several users compare it to Citymapper and Google/Apple Maps; in some cities competitors are still preferred.

Underground localization approaches discussed

  • Core idea: use accelerometer and motion classification to detect when a phone is on a moving train and infer position along known routes and schedules.
  • Alternatives and complements suggested:
    • BLE beacons or Wi‑Fi in tunnels and stations.
    • Cell-tower–based indoor positioning.
    • Microphone-based “sound signatures,” though seen as a privacy non‑starter.
    • Barometer/pressure changes when entering/exiting tunnels, used in older research; limited by inconsistent sensor availability.
    • Magnetometer/compass data and signals from in-train systems (CarPlay-style GPS sharing, Wi‑Fi APs broadcasting location).

Dead reckoning, sensors, and ML

  • Multiple comments note classical dead reckoning (integrating accelerometer/gyro) drifts quickly and needs absolute references.
  • Some suggest using distinctive acceleration “signatures” of each track segment, or “Shazam for train tracks.”
  • Others propose combining hunting oscillation of wheel-rail dynamics, inertial data, and schedules.
  • Debate over model types: CNNs, SVMs, RNN/LSTMs, and sensor fusion; one project member confirms they focused on accelerometer-based classifiers and did not use pressure sensors.

Accuracy, edge cases, and limitations

  • The team reports ~90% correct station prediction; commenters question whether that’s sufficient for trust.
  • Hard cases mentioned:
    • Trains stopping in tunnels, running slowly, or skipping stations.
    • Express vs local trains, wrong direction, and mixed-generation fleets.
    • Users walking on the train, packed cars, and faulty in-vehicle GPS.
  • Some users saw the app think they were at the wrong station for entire journeys.

Transit UX, ads, and information

  • Strong sentiment that onboard screens should prioritize next-stop info and door side, but often show ads or generic warnings instead.
  • Arguments that ad revenue matters for underfunded transit, but others say revenue is small and not worth degrading rider info.

Business model, privacy, and ecosystem

  • Transit monetizes via freemium features, in‑app ticketing, and agency partnerships; some dislike subscriptions, others find free tier sufficient.
  • Positive reactions to data staying on-device and not being used as a surveillance/marketing feed.
  • Some speculate this tech would be attractive for acquisition by larger platforms (e.g., Google), but this is not confirmed.