5 ms·

How exactly would you do that? It’s impossible to map from a dataset of non-unique identifiers to unique identifiers without additional data and heuristics. The

How exactly would you do that? It’s impossible to map from a dataset of non-unique identifiers to unique identifiers without additional data and heuristics. The mapping is ambiguous by definition. The underlying flight plan standard were all created in an era of low memory machines, and when humans were expected to directly interpret data exactly as the programs represented it internally (because serialisation and deserialisation is expensive when you need every CPU cycle just run your core algorithms)

zarzavat 3y ago
Couldn’t you use the surrounding points? Each point is surrounded by a set of nearby points. You can prepare a map of pairs of points into unique ids beforehand, then have a step that takes (before, current, after) for each point in the flight plan and finds the ID of current.
- seabass-labrax 3y ago
  That sort of thing happens already; for instance, the MCDU of an Airbus aircraft will present various options in the case of ambiguous input, with a distance in nautical miles for each option. Usually, the closest option is the most appropriate.
- avianlyric 3y ago
  Yes you can, but that would be using > additional data and heuristics Directly mapping is impossible, so you cant just do a dumb ID-at-time pre-processing step (which is what your comment seems to suggest). You need a more complex pre-processing step that’s capable of understanding the surrounding context the identifier is being used in. A major issue with the flight planing system (as highlighted in the article) is that they attempted to do this heuristic mapping as part of their core processing step, and just assumed the ID wouldn’t be too ambiguous, and certainly wouldn’t repeat.
codeulike 3y ago
If two waypoints have the same name assume its the closest one to the adjacent ones in the route-chain rather than the one 4000 km away
- moritzwarhier 3y ago
  Sounds like a helpful idea with considerable implementation complexity, including the potential for new disastrous failure modes.
  - zarzavat 3y ago
    Not really because each point is only adjacent to a small neighborhood of other points, so if you want to test every possibility then your search space only grows by a constant factor proportional to the maximum degree of the graph. As for implementation complexity, you would hope they would use formal verification for something like this.
  - codeulike 3y ago
    But its how the waypoint codes work in practice - they are contextual. If Air Traffic Control tell a plane to head for waypoint RESIA, they mean the one nearby, not the one 4000 km away.
    - moritzwarhier 3y ago
      Have to admit, I read the article in full detail only after commenting and I see your point. Especially since the implenting company is called out explicitly for failing to achieve this, and the risks of changing the well-established identifiers are also illustrated. Perfect might be the enemy of the good then, or the standardization thing at least is a separate topic.

zarzavat 3y ago

Couldn’t you use the surrounding points? Each point is surrounded by a set of nearby points. You can prepare a map of pairs of points into unique ids beforehand, then have a step that takes (before, current, after) for each point in the flight plan and finds the ID of current.

seabass-labrax 3y ago
That sort of thing happens already; for instance, the MCDU of an Airbus aircraft will present various options in the case of ambiguous input, with a distance in nautical miles for each option. Usually, the closest option is the most appropriate.
avianlyric 3y ago
Yes you can, but that would be using > additional data and heuristics Directly mapping is impossible, so you cant just do a dumb ID-at-time pre-processing step (which is what your comment seems to suggest). You need a more complex pre-processing step that’s capable of understanding the surrounding context the identifier is being used in. A major issue with the flight planing system (as highlighted in the article) is that they attempted to do this heuristic mapping as part of their core processing step, and just assumed the ID wouldn’t be too ambiguous, and certainly wouldn’t repeat.

codeulike 3y ago

If two waypoints have the same name assume its the closest one to the adjacent ones in the route-chain rather than the one 4000 km away

moritzwarhier 3y ago
Sounds like a helpful idea with considerable implementation complexity, including the potential for new disastrous failure modes.
- zarzavat 3y ago
  Not really because each point is only adjacent to a small neighborhood of other points, so if you want to test every possibility then your search space only grows by a constant factor proportional to the maximum degree of the graph. As for implementation complexity, you would hope they would use formal verification for something like this.
- codeulike 3y ago
  But its how the waypoint codes work in practice - they are contextual. If Air Traffic Control tell a plane to head for waypoint RESIA, they mean the one nearby, not the one 4000 km away.
  - moritzwarhier 3y ago
    Have to admit, I read the article in full detail only after commenting and I see your point. Especially since the implenting company is called out explicitly for failing to achieve this, and the risks of changing the well-established identifiers are also illustrated. Perfect might be the enemy of the good then, or the standardization thing at least is a separate topic.