This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
With messages as fresh as the radio allows, the next idea was to send what each craft intends, or how it is accelerating, so that listeners could predict past the delay.
What was tested
Two predictions were carried on the message: the speed and direction a craft is commanding, and its current acceleration.
What was found then
Intentions helped one test (4 collisions to 1) but hurt the streams (9 to 23), because a craft mid-dodge 'intends' to go back to its lane. The acceleration prediction looked good until the 84 slow-link cases (6 to 10), and a four-layout rerun confirmed the loss (16 to 30). When the last message can be seconds old, the manoeuvre it describes has usually ended, so extending it costs more than it saves. The plain straight-line guess stays.
What changed
Nothing (both predictions rejected).
Numbers and source references
- 4 → 1
- airspace twelve seeds with intentREADME.md:659
- 9 → 23
- five-family turbulence set with intentREADME.md:659
- 84
- cliff casesREADME.md:659
- 6 → 10
- cliff set with the acceleration predictorREADME.md:659; git 6b06f18
- 16 → 30
- four-seed rerun of the moved rowsREADME.md:659