This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
The crossing and three-stream families still had 22 collisions over sixteen runs in 3 m/s turbulence.
What was tested
We read the last eight seconds before each collision. Then we added a gust margin: the predicted miss distance is reduced by how far gusts could push each craft before the closest point.
What was found then
The dodge was ordered, but speed-matching and the pull toward the goal weakened it. And the prediction trusted the current gust, while gusts change within about 2.5 s. With the gust margin, which is zero in still air, stream collisions fell from 22 to 1. Pad test runs went from 9 to 0 with the same landings. The full test set (every scenario family and variant, about 400 cases) fell from 21 to 7. Every still-air result stayed exactly the same, and the work score (trips and cargo delivered) fell only 0.14 %.
What changed
A gust margin in the predicted closest approach.
Numbers and source references
- 22
- stream collisions before the fixREADME.md:164
- sixteen
- 1,200 s runs in the stream batteryREADME.md:164
- 3 m/s
- turbulence levelREADME.md:164
- 2.5 s
- gust memoryREADME.md:164
- 22 → 1
- stream batteryREADME.md:164
- 9 → 0
- pad probe batteryREADME.md:164
- 21 → 7
- full-matrix collisions (run 12)README.md:164; git 0db08a2
- 0.14 %
- less workREADME.md:164