This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
The remaining collisions were pairs at the same height, each heard 2–18 m from where it really was, on news 2.1–5.6 s old. No steering rule can fix a position nobody has heard.
What was tested
A craft whose nearest heard neighbour is within two required spacings now sends four times as often, but never more than once every 0.25 s.
What was found then
The first version overflowed the queue of delayed messages and made one turbulent test worse (9 collisions to 18). With a bigger queue, the slow-link cases (a link just too slow and lossy to keep up) fell from 13 to 5, with work up 0.8 %. In a twelve-layout crowded-sky test, four impacts were left. Their news was 2.1–3.1 s old, as fresh as a 2 s delay allows. The full test set stayed at 0.
What changed
Extra messages when a neighbour is close, at most one every 0.25 s.
Numbers and source references
- 2–18 m
- error in the heard positionREADME.md:615
- 2.1–5.6 s
- message ages at impact beforeREADME.md:615
- two standards
- distance that triggers the faster sendsREADME.md:615
- 0.25 s
- fastest send intervalREADME.md:615
- 9 → 18
- turbulent set with the overflowing first buildREADME.md:615
- 13 → 5
- cliff setREADME.md:615; git 4df11a8
- +0.8 %
- workREADME.md:615
- 2.1–3.1 s
- message ages at the remaining impactsREADME.md:615
- 2 s
- link delayREADME.md:615
- −0.06 %
- work on the full matrix (run 21)README.md:615
- four
- impacts left on the twelve-seed airspace setREADME.md:615