evidoria

← Back to browse

Good practice Imported

TrackInspect — the MTA and Google Public Sector's AI Pilot for Predictive Subway Track Maintenance

United States of America · New York · See the United States of America profile · See the New York profile

Evidence: Observational / pre–post Top 83% 40/100 · Ask Evidence Copilot about this practice

New York's MTA retrofitted Google Pixel phones onto six subway cars to capture vibration and audio data, feeding an AI model that matched 92% of the track defects human inspectors found over a five-month Google Public Sector pilot on the A line.

335 million
Sensor readings collected (Sep 2024-Jan 2025)
1 million
GPS points collected (Sep 2024-Jan 2025)
1,200
Audio hours collected (Sep 2024-Jan 2025)
92 %
Match rate with human-inspector-found defects
TrackInspect — the MTA and Google Public Sector's AI Pilot for Predictive Subway Track Maintenance

Details

Maturity
Pilot
Promoter
Metropolitan Transportation Authority (MTA) / Google Public Sector
Period
September 2024 – January 2025 pilot
Keywords
public transit, infrastructure maintenance, predictive maintenance, transportation

Context

The MTA (Metropolitan Transportation Authority), which runs the largest subway system in North America, faces a constant backlog of manual track inspection. In September 2024, New York City Transit and Google Public Sector launched TrackInspect, retrofitting six R46 subway cars on the A line with off-the-shelf Google Pixel smartphones mounted to capture vibration and audio signals as trains ran their normal routes.

Activities

Over the pilot's five months (September 2024-January 2025), the phones collected 335 million sensor readings, 1 million GPS points and 1,200 hours of audio, which cloud-based machine-learning models analyzed to flag likely track defects such as loose fasteners or rail irregularities. Human track inspectors then verified each flagged location and fed their findings back into the model.

Results

TrackInspect's flagged locations matched 92% of the defects independently found by MTA's human inspectors.

Conclusions

Track inspectors remain 'in the loop': the system flags candidate locations for human confirmation rather than acting autonomously, and its performance figures come from MTA and Google's own reporting rather than an independent third-party audit. The pilot covered only six cars on a single line; MTA has stated an intention to expand AI-assisted inspection system-wide, but as of the pilot's public reporting no confirmed system-wide rollout, budget commitment, or post-pilot funding arrangement with Google Public Sector (which developed the tool at no cost during the pilot) had been disclosed.

Implementation
Indicative cost
Low (< €50k) — Developed at no cost to MTA by Google Public Sector during the pilot; no post-pilot funding arrangement disclosed.
Time to results
Short (< 1 year) — Pilot ran September 2024-January 2025.
Staffing & skills
MTA/New York City Transit track inspectors, Google Public Sector engineering team

Conditions for success

  • Keeping human track inspectors 'in the loop' to verify flagged locations rather than acting autonomously
  • Using off-the-shelf smartphones instead of custom hardware to keep the pilot low-cost

Common failure modes

  • Performance figures come from MTA and Google's own reporting rather than independent third-party audit
  • No confirmed system-wide rollout, budget commitment, or post-pilot funding arrangement disclosed

Where it fits

Governance type
public transit authority + private technology vendor pilot
Scale
single line, six cars (New York City)
Income level
high-income

Commonly funded by

Own resources / municipal budget

Indicative funding routes for practices of this type — always check each programme's current calls and eligibility rules.

Do you run this practice? Claim it — verified implementers get a public contact pathway and can propose corrections.

Data sources

Where this practice's information was retrieved from, and when.

Attachments

Similar practices you may find useful