Units were ordered to move from left to right (quite a long distance actually). They didn't manage to steer clear of the obstacle. Instead they got stuck multiple times in the locations were i added the clock symbols, and it always took them 1-2 seconds to recognize that they got stuck before they would start moving around the obstacle. The next batch of units would get stuck at the same location again. (Despite having a common movement goal!) The units at the location of the lower icon were actually already stuck at the middle icon before, so they ran straight into the next deadlock after escaping the first one. Actual target was reachable the entire time. Observable in Lobby 16230637765778171451 at mark 60:30 to 62:00
Sorry, for push, but I think I do understand now what is causing the issues. First of all, this issue only appears to occur with platoons, not with single units. Platoons which are moving in formation, or to put it better: Which are forced into formation by the flocking algorithm. It appears as if the flocking algorithm is weighted higher than direction changes in the vector field, which results in units following the direction dictated by the flock, rather than avoiding obstacles as the vector field would have told them too. There is also a second bug somewhere in the generation process which apparently causes the flooding algorithm to traverse nodes in the wrong order, resulting obstacles not properly propagating, but that's an entirely different issue.
Try latest PTE. There is a fix in that version that explicitly tries to alleviate this situation. It won't make every situation better, but you should see significantly less of it.
I'm on PTE 71609 right now, and it's still happening on a regular base (almost 100% reproducible). At least the "getting stuck" part. For platoons which were supposed to avoid obstacles, only 1/2 of each platoon does so. The other half runs right into it. And if the obstacle happens to have an indention, units get stuck and become unable to move out of the indention for unknown reason.
Unfortunately not fixed in 71729 either. It's the flocking system + physics pushing units into these locations in the first place, causing units to slide into the "forbidden" location. But once they are colliding, they act like the obstacle doesn't exist at all? Uhm, could it be that the magnitude of the vectors in the pathing field is clipping/overflowing right next to impassable obstacles, so it's no longer pointing into any valid direction? Causing units to fall back to default bias, which is probably "direction of target"? And that the area where the vector field has invalid values is just slightly bigger than the collision box used by the physic subsystem?
