This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Lanes should keep craft in a 10 m-tall box. Two earlier ways failed where lanes cross at one height. Last try: a dodge planner (the part choosing each craft's avoiding move) that never dodges up or down inside the box.
What was tested
Six ten-minute layouts on a grid of same-height crossings, with ideal craft (they respond instantly) and slow-responding test craft, and a busy mixed sky at four times traffic; then we measured where craft fly.
What was found then
Ideal craft went from 0 to 4.4 collisions per 1000 km, slow craft from 3.9 to 80, the busy sky nearly doubled, and close calls became near-constant. Craft normally fly a band about ±12 m tall and spend only 20–25 % of the time in their box. A reading that it stopped slow lanes sinking was taken back in cycle 14: collisions moved up, not lanes.
What we learned later
The ±12 m band turned out to be where the gentle pull back toward the lane holds craft, not room needed for dodging.
research/reviews/brief-c19.md
What changed
Nothing. The planner stayed a test switch, off by default.
Numbers and source references
- 6 × 600 s
- seeds × run length per caseresearch/reviews/brief-c13.md
- 4×
- airspace traffic multipleresearch/reviews/brief-c13.md
- 0 → 4.4
- ideal craft collisions per 1000 km, default vs in-band planner (4.37)research/reviews/brief-c13.md
- 12 % → 96 %
- run time with some pair inside the separation standard, ideal craftresearch/reviews/brief-c13.md
- 3.9 → 80
- slow craft collisions per 1000 km (3.93 → 79.6)research/reviews/brief-c13.md
- 3.9 → 7.3
- busy airspace ×4, slow craft (3.93 → 7.29)research/reviews/brief-c13.md
- ±12 m
- vertical spread (p10/p90) of ideal craft around the lane centreresearch/reviews/brief-c15.md
- 20–25 % / 10–13 %
- time in the 10 m box, ideal / slow craftresearch/reviews/brief-c15.md
- 22 m / 39 m
- median vertical gap of horizontally-near lane-mates, ideal / slowresearch/reviews/brief-c15.md