This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Collisions between lanes had been analysed as if craft stayed in their lanes.
What was tested
We measured sideways straying from the lane, added a gentle pull back into the lane, counted height errors more in the extra-message trigger, and told craft in crossing lanes which way to dodge vertically, from the lane layout.
What was found then
Craft were not in their lanes: the typical stray was 414 m in still air, on a 240 m lane width. With the pull back it was about 90 m in turbulence. Counting height errors properly cut one four-layout test from 12 collisions to 4, for about a tenth more messages. With the lower lane dodging down and the upper up, lane-against-lane collisions fell from 9 to 0. Sending intended height instead was rejected (dense pad collisions 0 to 14). The full test set stayed at 0.
What changed
A pull back into the lane (gentle sideways, stronger vertically, capped at 0.35 of cruise speed); height errors weigh more in the extra-message trigger; dodge direction taken from the lane layout.
Numbers and source references
- 414 m
- median lateral scatter beforeREADME.md:544
- 240 m
- lane widthREADME.md:544
- 90 m
- median scatter after, in turbulence 3 (about)README.md:544
- 12 → 4
- four-seed airspace set with height-weighted triggerREADME.md:565
- a tenth
- more messages (+9.5 %)README.md:565
- 9 → 0
- lane-versus-lane collisionsREADME.md:565
- 0 → 14
- dense terminal with intent on the message (rejected)README.md:565
- +0.02 %
- work (run 20)README.md:565
- 0.1 /s
- lateral pull-back gainREADME.md:544
- 0.3 /s
- vertical pull-back gainREADME.md:544
- 0.35
- pull-back cap as a share of cruiseREADME.md:544