No luck, sorry. The area looks clear from any obstacles or stray polygons in my data.
Some background:
AC3 does have invisible polygons in some places; I mentioned it
here. These polygons use blank texture areas and are therefore completely invisible when rendered with texture mapping. Out of pure luck, early versions of my datasets did not decode the texture positions properly, and so the polygons became visible. Here they are for
Tunnel Vision:
In the same dataset, I cannot spot anything suspicious in
Claustrophobia:
(I tried all angles, of course!)
Then I fired up my engine (which has collision detection) and checked if it detects collisions even with those invisible polygons. And it does – flying next to the tunnel in
Tunnel Vision, towards the blocks from the first image, rings my terrain collision warning and then makes my plane crash into the invisible wall:
But nothing like that happens in
Claustrophobia. I have flown the path up and down, multiple times; circled there and chased the canyons … but I never collided with anything :(
Here is my guess:
If you look at the crash sites, they seem very close to the seams where terrain tiles meet.
I haven’t found a spatial structure in AC3’s levels (yet?). There is no BSP tree or anything alike to speed up collision detection, just lists of polygons. So I don’t know how the game tests for collisions. But should these spurious crashes indeed happen on terrain tile edges, then the most likely explanation is an overrun in the fixed-point coordinates during collision detection – i.e. a software bug. Ray vs. Terrain is hard in fixed-point, and even more so with fast planes and small tiles like AC3’s.
Claustrophobia has the most chaotic terrain with the highest cliffs, so the error would most often be observed there (and maybe in
Ghosts of the Past).
tl;dr: Don’t know, sorry. This is an OG bug for real.
