You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
{{ message }}
Repository navigation
TrailView features: the road solver's stop rule can leave a job metres from its minimum #3463
Part of #3096. Follow-up of #3449 (the road height solver that seeks the minimum of its rules).
Problem
The solver stops a job when no vertex alone would move by 1 cm (SETTLE_M) or at the round cap. On a long stiff deck this can leave the heights metres from the minimum of the cost. The result is deterministic, but it is not the minimum, and it depends on where the first scale stopped.
Measured by the last review of #3449 on the Seelisberg corridor (solver-fix2/delta3-review/ on timosPC):
With 300 rounds at scale 1/100 in place of 40, the result at scale 1 moves by up to 12.0 m on the A2 deck (309 vertices by more than 1 m).
With SETTLE_M = 1e-4, both starts agree within 0.08 m. The minimum is unique. The merged head's corridor result lies up to 17.8 m from it.
The tight rule takes the roads step of that corridor from 1.0 s to 3.3 s.
On test box B, 2 jobs still end at the round cap at scale 1/100 (last move 0.47 m). The cause of their 2-cycle (one sag row of a bridge chain that flips by about 5e-5 m) is not found.
Wanted
A stop rule that bounds the distance to the minimum: for example on the joint step, or on the fall of the cost, in place of the move of one vertex. The whole-box effect and the time must be measured with one box bake each (5 GB cap). The owner's rule on bake cost: growth is fine for a real gain while it stays far from several times; report growth over a few percent.
Activity