Show HN: Agentic Airport (github.com)

3 points by ovisource ↗ HN
A browser based air traffic control simulation where agentic AI acts as a control tower with an objective of landing airplanes.

It was developed as an experiment in exploring agentic AI capabilities for controlling multiple objects in an active space.

The results were spectacular as a single agent can not only land multiple airplanes, but it also prevents collisions with other airplanes in small busy spaces.

1 comment

[ 3.7 ms ] story [ 13.5 ms ] thread
Former ATC here ..

This is a fun experiment, but not very realistic in its current form. Handling of traffic is based on two principles. 1. Safety, 2. Efficiency.

1. Safety

In essence, aircraft separation can be either vertical (usually 1000ft) and/or horizontal (usually 5 nautical miles) .. There are variations of this, depending on radar coverage (can accept less horizontal separation with better coverage) and wake turbulence categories (large, powerful planes cause turbulence that can affect smaller planes or helicopters - more spacing can be required between a large and a small aircraft).

2. Efficiency

To be effective means not only to get all flights on the ground as soon as possible in an orderly manner, but also for the individual flight to be as cost effective as possible. What is most cost effective? Not using fuel. All commercial aircraft are built to stay in the air - and when using idle power (engines not producing significant thrust), to glide for long distances.

So all flights should follow an optimal three dimensional curve, that requires: 1. Optimal distance: To leave their planned maximum altitude for the route by simply reducing thrust, making the aircraft descend, at the optimal point where the aircraft can simply glide all the way to the runway without using additional thrust (fuel). 2. Optimal path: To minimise the amount of turns required, which causes drag, which causes the aircraft to descend faster (would have to use thrust at some point) 3. Optimal speed: To ensure the aircraft touches down on the runway with the required speed.

---

An Air Traffic Controller has been trained to consider such factors, but also has the benefit of "the airspace system". Controlled airspace has predefined routes (highways in the air) which are crafted based on aircraft operational conditions. On arrival, aircraft follow a "STAR", which is a predefined route with predefined altitude requirements (cross point X with altitude Y) - usually a mix of an optimal glidepath for the type of aircraft the airport serves, surrounding airspace limitations (restricted airspace, military areas) and geographic features (nice to avoid mountains).

So to facilitate the flow of traffic, the ATC depends on using the airspace as it is designed - and deviating from that when necessary. Deviations are usually called "vectors". The ATC will instruct a flight to fly a certain heading (based on radials of 360 degrees, extending from the flight itself) .. In general "Fly heading 360" (north), "Fly heading 090" (east), "Fly heading 180" (south).. etc or more fine grained - "Fly heading 045" (north-easterly). A deviation can also be a certain flight level, or altitude. Due to traffic ATC can request a flight to descend earlier than planned, or to cross a point on a route at a certain altitude.

An air traffic controller will, traffic permitting, try to avoid deviations from the aircrafts planned route, since the planned route is the "optimal" trajectory for the flight (although constrained by airspace design). Also - placing aircraft on vectors reduces the ATCs mental capacity (the ATC has to remember to give further instructions at a later point, for the flight to return to its planned route - or some other instruction). So an ATC would try to minimise the number of flights on vectors. Unless he/she is doing approach sequencing.

---

Approach sequencing.

So at the final stages of the flight, approaching the airport - the flights should as far as possible mostly following predefined routes. These routes end up going alongside the runway (in the opposite direction), and then "U-turning" into the runway direction with a nice smooth curve.

Think of a straight line extending out in the runway direction - guiding aircraft toward the runway (like the green dots you have on your example). This is the function of the localizer (