AI Takes the A-Train and Also Goes Under the Sea

While subjects such as OpenAI, ChatGPT, and DeepSeek grab the headlines, practical applications are often subtle in their impact, take time to develop, and may get lost in the background noise. The examples provided below illustrate the growing importance of AI. They also show that human ingenuity, work, and time are required to develop important applications.

TrackInspect Pilot

Recently, the Rapid Innovation Team at Google Public Center and the New York Metropolitan Transit Authority (MTA) released a report on “TrackInspect,” a pilot proof-of-concept program that uses AI for the early detection of possible track problems. The pilot project was tested on the New York City subway’s A-line, which was made famous by the Duke Ellington Band’s 1939 song “Take the A-train” and operates around the clock and carries 600,000 passengers per day. 

The MTA, like many railroads, uses specialized “track geometry cars” (TGCs) to identify anomalies in tracks. These are non-revenue cars, i.e., they do not carry passengers or freight, are operated by dedicated personnel, and use lasers and audio sensors to measure track and right-of-way features. In general, TCGs are designed to run on a non-interference basis with normal rail traffic, but because of the frequency service, i.e., 2 to 5 minutes apart during rush hours on New York City subway lines, they usually operate only at night or on weekends.

The “TrackInspect” pilot, on the other hand, used standard Google Pixel smartphones with off-the-shelf plastic cases attached to MTA’s passenger subway cars.  Hence, it can collect more data on a 24-hour/day basis. During the pilot project, data from 335 million sensor readings, one million GPS locations, and 1,200 hours of audio was collected. Sound and vibration data was sent in real-time to cloud-based systems, where AI and machine learning algorithms were used to generate predictive insights. Track inspectors served as “humans in the loop,” inspecting locations highlighted by the system, confirming whether there was an issue, and providing feedback to continuously train the model. TrackInspect also utilized Generative AI for natural language processing, allowing inspectors to ask questions about maintenance history, protocols, and repair standards, with clear, conversational answers.

(Source: New York City Transit)

Based upon the experience gained during the pilot program, the MTA has issued a Request for Expressions of Interest from firms with expertise in sensor deployment, data collection, and/or AI/ML-driven analytics with the goal of developing and deploying a subway system-wide continuous real-time data collection and actionable analysis system to supplement existing TGC and manual inspections. 

By retrofitting passenger cars with commodity hardware, the MTA aims to create a continuous and scalable monitoring system that supplements traditional inspection methods, enabling earlier detection and proactive maintenance providing for more reliable service as well as reducing maintenance and operational costs.

The subway system has 6,712 passenger cars and carries 1.6 million passengers a day. Implementation on a system-wide basis would obviously result in a collection of a vast amount of information for real-time analysis.

Separately, the New York City Transit/MTA Bus (NYCT/MTAB) agencies released a request for information seeking to identify potential sources for developing, delivering, and maintaining a similar Vehicle Telematics and Data Analytics System (VTaDAS) for MTA’s zero-emissions bus operations.

The NYCT/MTAB operates a bus network spanning all five boroughs of New York City, comprising around 6,000 buses. The service runs 234 local, 71 express, and 20 Select Bus routes continuously, 24/7, carrying 1.6 million passengers daily on average, and covering approximately 120 million miles annually.

Through VTDAS the NYCT/MTAB expects to improve the reliability of the bus network by collecting and processing in real-time vehicle operational status data including trip information, energy consumption, charging history, operating parameters, driver behavior, diagnostic messages, mechanical fault data, and passenger data.

Underwater Mining and Intelligent Disposal of Discarded WW2 Munitions

(Source: archy13/Shutterstock)

It is estimated that 1.6 million tons of explosives and chemical weapons were discarded in German waters alone at the end of World War II. At the time, this was considered a safe disposal method. In the 80 years since the end of the war, there have been numerous injuries and fatalities from these discarded munitions. In addition, ongoing monitoring of the waters has shown that chemicals leaching out of these abandoned munitions have found their way into the food chain. It is expected that as the casings of the munitions deteriorate, food chain contamination will worsen.

In 2024, the German government funded two companies, SeaTerra and Eggers Kampfmittelbergung, to develop the technology to remotely scan for and identify the type of munition and its condition. This information is needed to recover and dispose of these abandoned munitions.

Currently, the project is scanning munitions and creating a machine learning database. The longer-term goal is to use AI and robots to find, identify, retrieve, and properly dispose of the munitions in floating incineration facilities.

About the Author

Paul Muzio is currently an advisor to Intersect360’s HPC AI Leadership Organization (HALO). Previously, he held positions in HPC at the City University of New York, Network Computing Systems, Inc., and Grumman Aerospace Corp.