Reading List 11/8/2025
Gathering robot training data, “love letters” sent to home sellers, the Napier Deltic diesel engine, jumps in electricity demand from electric teakettles, and more.
Welcome to the reading list, a weekly roundup of news and links related to buildings, infrastructure, and industrial technology. This week we look at gathering robot training data, “love letters” sent to home sellers, the Napier Deltic diesel engine, jumps in electricity demand from electric teakettles, and more. Roughly 2/3rds of the reading list is paywalled, so for full access become a paid subscriber.
Robot training
We’ve previously noted that one major bottleneck in making robots more capable is a lack of training data. LLMs have the benefit of the entire internet as a source of training data, but there’s no such pre-existing “movement” dataset that we can use to train robot AI models on, and finding/creating a source of robot training data has become an important aspect of robot progress. The LA Times has a good piece about some of the companies working to collect this training data:
In an industrial town in southern India, Naveen Kumar, 28, stands at his desk and starts his job for the day: folding hand towels hundreds of times, as precisely as possible.
He doesn’t work at a hotel; he works for a startup that creates physical data used to train AI.
He mounts a GoPro camera to his forehead and follows a regimented list of hand movements to capture exact point-of-view footage of how a human folds.
That day, he had to pick up each towel from a basket on the right side of his desk, using only his right hand, shake the towel straight using both hands, then fold it neatly three times. Then he had to put each folded towel in the left corner of the desk.
If it takes more than a minute or he misses any steps, he has to start over.
Privatized air traffic control
In response to the ongoing government shutdown, an air traffic controller shortage is forcing a curtailment of airline flights across the US. Air traffic controllers are having to work without pay, and there are about 3500 fewer controllers working than needed. On Thursday 10% of flights across the country, around 1800 flights, were ordered to be cancelled.
Air traffic control seems like an obvious service for the government to provide, like police or firefighters, but apparently private or semi-private air traffic control systems aren’t all that uncommon. Marginal Revolution has an interesting, short post about Canada’s privatized air traffic control system:
It’s absurd that a mission‑critical service is financed by annual appropriations subject to political failure. We need to remove the politics.
Canada fixed this in 1996 by spinning off air navigation services to NAV CANADA, a private, non‑profit utility funded by user fees, not taxes. Safety regulation stayed with the government; operations moved to a professionally governed, bond‑financed utility with multi‑year budgets. NAV Canada has been instrumental in moving Canada to more accurate and safer satellite-based navigation, rather than relying on ground-based radar as in the US.
“NAV CANADA – in conjunction with the United Kingdom’s NATS – was the first in the world to deploy space-based ADS-B, by implementing it in 2019 over the North Atlantic, the world’s busiest oceanic airspace.
NAV CANADA was also the first air navigation service provider worldwide to implement space-based ADS-B in its domestic airspace.”
Meanwhile, America’s NextGen has delivered a fraction of promised benefits, years late and over budget. As the Office of Inspector General reports:
“Lengthy delays and cost growth have been a recurring feature of FAA’s modernization efforts through the course of NextGen’s over 20-year lifespan. FAA faced significant challenges throughout NextGen’s development and implementation phases that resulted in delaying or reducing benefits and delivering fewer capabilities than expected. While NextGen programs and capabilities have delivered some benefits in the form of more efficient air traffic management and reduced flight delays and airline operating costs, as of December 2024, FAA had achieved only about 16 percent of NextGen’s total expected benefits.”
NPR also ran an article in July this year about the debate around switching to a privatized system in the US, and the pros and cons of a system like Canada’s:
Canada went from paying for air traffic control largely through tax revenue to charging customers a fee based on the weight and distance of a flight.
According to Correia, privatizing air traffic control was the next move for an aviation sector that already had privately-held airplane manufacturers and commercial airlines. “So basically the step that was taken by Canada was to say, well, air traffic control is providing a service to an industry that is already privatized or mostly privatized in many regions of the world,” he said.
Other air traffic control systems that exist outside or partially outside the government include NATS in the United Kingdom, Airservices Australia, Airways New Zealand, DFS in Germany and Skyguide in Switzerland.
A 2017 report by the Congressional Research Service said other countries’ models don’t appear to show “conclusive evidence that any of these models is either superior or inferior to others or to existing government-run air traffic services, including FAA, with respect to productivity, cost-effectiveness, service quality, and safety and security.”
The origins of Airbus
I’ve previously written about the difficulties of competing in the commercial aircraft industry: the costs of developing commercial aircraft are so high (often a significant fraction of the value of the company) and the number of annual aircraft sold so few, that a few bad bets — a program that goes over budget, or sells much less than anticipated — can be ruinous.
Given these difficulties, and the fact that many companies (Lockheed, Douglass, Convair) have been forced from the field, it’s a little surprising who ended up being competitive, and who didn’t. Japan, despite overwhelming many US industries with inexpensive, high-quality manufacturing in the second half of the 20th century, never fielded a commercial airliner (though not for lack of trying). South Korea didn’t either. Instead, the international competitors came from Brazil (Embraer), and a consortium of European countries (Airbus). Works in Progress has a good piece on why Airbus was so successful, when so many other similar European efforts failed:
Europe is a graveyard of failed national champions. They span from the glamorous Concorde to obscure ventures like pan-European computer consortium Unidata or notorious Franco-German search engine Quaero.
Airbus is the rare success story. European governments pooled resources and subsidized their champion aggressively to face down a titan of American capitalism in a strategically vital sector. Why did Airbus succeed when so many similar initiatives crashed and burned?
Airbus prevailed because it was the least European version of a European industrial strategy project ever. It put its customer first, was uninterested in being seen as European, had leadership willing to risk political blowback in the pursuit of a good product, and operated in a unique industry.
…Roger Béteille, who led the A300 program, probably bears more responsibility for Airbus’s early success than anyone else. Béteille wasn’t interested in building an inferior European Boeing copy. Instead, he invested significant time in getting to know his potential customers and what they needed. This led to Airbus quickly tossing the original design for a 300-seat A300, in favor of a 225-250 seater, when it became clear that Air France and Lufthansa wanted a smaller product.
The revised A300B would prove much cheaper to develop, in part because it allowed the consortium to dispense with the expensive Rolls Royce engine in favour of a cheaper American alternative. In response, the UK exited the project, only to later return with a lower ownership stake.
This willingness to risk political blowback and avoid petty chauvinism in equipment choice was rare in industrial strategy.
Béteille went one step further. He designated English the official language of the project, instead of the usual mixture of languages that characterised European projects, and forbade the use of metric measurements to make it easier to sell into the US market.
Air travel computational complexity
Also on the subject of commercial air travel, air travel reservation systems, like early spell checkers, are one of those pieces of software that’s much more complicated than it might first appear. One of the first major air travel reservation software systems, SABRE, was built by IBM using the technology from the recently-completed SAGE defense system (one of the most expensive megaprojects ever built). And these 2003 lecture slides from an MIT course on artificial intelligence talk about some of the computational difficulties involved in booking flights:
At 30,000,000 flights per year, standard algorithms like Dijkstra’s are perfectly capable of finding the shortest path. However, as with any well-connected graph, the number of possible paths grows exponentially with the duration or length one considers. Just for San Francisco to Boston, arriving the same day, there are close to 30,000 flight combinations, more flying from east to west (because of the longer day) or if one considers neighboring airports. Most of these paths are of length 2 or 3 (the ten or so 6-hour non-stops don’t visually register on the chart to the right). For a traveler willing to arrive the next day the number of possibilities more than squares, to more than 1 billion one-way paths. And that’s for two airports that are relatively close. Considering international airport pairs where the shortest route may be 5 or 6 flights there may be more than 1015 options within a small factor of the optimal.
One important consequence of these numbers is that there is no way to enumerate all the plausible one-way flight combinations for many queries, and the (approximately squared) number of round-trip flight combinations makes it impossible to explicitly consider, or present, all options that a traveler might be interested in for almost all queries.