This is an earlier experiment. Read the amendments below before treating its findings as current.
The question
Until now every craft had perfect information. Real craft hear each other over a radio with limited message rate, lost messages and delay.
What was tested
We added a link model with rate, loss and delay. Between messages, a craft assumes a neighbour kept flying straight at its last heard speed. Four worse links joined the full test set, and harsher links were tried on a second machine.
What was found then
The first 56 worse-link cases had one collision. A harder 56-case run at 2 messages a second, 30 % loss and 1 s delay had none. The breaking point was 1 message a second, 50 % loss and 2 s delay. There, news can be over 3.7 s old: long enough for a neighbour to move a whole required spacing. A fast, lossless link with no delay reproduced perfect information exactly.
What changed
Link model added (off by default).
Numbers and source references
- 2 messages a second, 30 % loss and 1 s delay
- link that heldREADME.md (Stress campaign) (The link)
- 1 message a second, 50 % loss and 2 s delay
- the cliffREADME.md (Stress campaign) (The link)
- 3.7 s
- message age beyond which a neighbour can move a standardREADME.md (Stress campaign) (The link)
- 20 Hz
- lossless, zero-latency link that is bit-identical to perfect informationREADME.md@e25c970:583
- Four
- degraded links added to the matrixREADME.md (Stress campaign) (The link)
- 56
- cases in each of the two link runsREADME.md:344
- one collision
- in the first link runREADME.md:344