road tracing

[joevenzon]

seems to work but the new roads.txt is almost identical to the old one (i just updated ring2007).

Is the actual behavior in-game as you expect?

[alex25]

[Is the actual behavior in-game as you expect?

i am not really sure how to tell the difference. the car seems to behave somewhat differently from before, but that's just about it. i did expect something else but i'm afraid i am not able to quite quantify that.

--alex--

[joevenzon]

I assume at least the car's wheels should be at the proper height on roads that have a lot of curvature in the left-right direction (crowning) to them... does that work OK?

[alex25]

I assume at least the car's wheels should be at the proper height on roads that have a lot of curvature in the left-right direction (crowning) to them... does that work OK?

finally managed to find a test case. the short answer is: NO, the new road tracing doesn't behave as i would expect it. look at these pictures:




the car seems to sit on a plane that connects the edges of the road (which is underneath the actual road) instead of following the road surface (at this particular location. each inflection point in the road corresponds to a tracing point. there should be 4 points).

--alex--

[joevenzon]

Okay, so a couple of things:

1) I thought the desired outcome was for the road to have a smooth, continuous curve from left to right that follows the contours of say a road that is bowed out so the rain drains off. Based on the screenshot it seems you were expecting it to actually be three or four discontinuous segments from right to left. This is hard to describe, but do you follow what I'm saying? What I did was make the track editor allow you to place the middle control points for the bezier surface, which should make it curved a bit in the direction of the control points.

2) The control points probably aren't having a large enough effect, so I'd need to tweak that some to get it right. But maybe this isn't necessary if it's not the effect you were interested in and I need to go back to the drawing board anyway.

[alex25]

I thought the desired outcome was for the road to have a smooth, continuous curve from left to right that follows the contours of say a road that is bowed out so the rain drains off. Based on the screenshot it seems you were expecting it to actually be three or four discontinuous segments from right to left.

indeed, i was expecting multiple segments which i think would describe the road more accurately.

--alex--

[joevenzon]

Okay, to make that happen I need to rethink my approach. How much work was the extra tracing you just did? Ideally a different approach would change how the track editor works when in 3 and 4 segment mode, but then that'd make it so you'd have to re-trace any tracks you updated with 3/4 segments. I might be able to make VDrift work with the data it already has from what you've done, but it's a less clean solution.

[alex25]

Okay, to make that happen I need to rethink my approach. How much work was the extra tracing you just did? Ideally a different approach would change how the track editor works when in 3 and 4 segment mode, but then that'd make it so you'd have to re-trace any tracks you updated with 3/4 segments. I might be able to make VDrift work with the data it already has from what you've done, but it's a less clean solution.

it's not that much work. for the ring i have a list of just the track segments so redoing the road tracing will be pretty fast. the other tracks are a lot easier to do and it won't take that much time anyway.

--alex--

[alex25]

hey joe, just wanted to make sure this doesn't get lost. are you still thinking about fixing the road tracing?

--alex--

[joevenzon]

I haven't forgotten it, but I'll add it to the issue tracker to make sure I don't.

I was thinking about this a bit more. In the test case of the nurburgring, what you want is one road segment that the AI follows, and then additional road segments that allow for a smooth road but which the AI won't take. For that, I think you could just trace the main road segment along the path you want the AI to take and then trace completely separate road segments for the other areas. Are there other cases where this approach doesn't work? What are those cases and what is the expected behavior?

[alex25]

looking at the pictures i posted above, and as i stated in a previous message, for that particular road segment i expected three discontinuous track segments. presumably they are all valid, if the car is going slowly (the engine blew out, or there is rain) i would expect the ai car to take the upper one (flat, to the left in the picture). normally the car would go on the bank segment in the middle. if you look at the very first picture i posted in this thread, that road is made up of two discontinuous segments (both flat but there is an angle they make with respect to each other). which is the correct one? personally i don't really think we need to smooth out the tracks, the number of polygons has increased dramatically and actually track authors put bumps and such on the created tracks to accurately represent the real tracks and vdrift just smooths them out. maybe we just need to come up with a way to tell the ai where the track boundaries are but oherwise just follow the actual track segments. does any of this make sense?

--alex--

[joevenzon]

Yes, this makes sense. So you need to be able to define separate road segments and then link them in some way so the game logic treats the similarly.

[alex25]

would it be possible just to trace the track. i.e. have one line for the right edge of the track and another for the left one. this would mark the track. have the AI work out the best path based on this information. other than that you just drive on the triangles that make up the track which are marked collide-able.

actually we still have the drive-able flag. maybe we should go back to using it marking the track, pit lanes, etc. drive-able. change the track editor to drive-able parts in a slightly different colour and allow you to pick triangles to mark/unmark drive-able (instead of picking vertices as it does now). what do you think?

[joevenzon]

The only complication is that the bezier patches which are generated from the "traced" data (and which make the track smooth to drive on) are quads. It'd take some work to take the triangle mesh and algorithmically determine which pairs of triangles make up the quads. The "auto-trace" feature in the track editor is an example of that sort of an algorithm and as you've probably noticed it screws up from time to time, especially on complex situations like with roads joining or branching to/from the track.

The easiest way would be to do it the same way as now with the road tracing, but add a way to mark different road segments as part of the same road (and maybe others as pit lanes or something) and have the AI figure out the rest.

[alex25]

just for fun, i removed roads.trk file for ring2007 and did one lap just to see how the car behaves. i don't know, it did feel somewhat more realistic. there are a lot of little bumps here and there (put in on purpose) which get smoothed out by the track segments. didn't see any major problems, but the second time around going over a crest the car just took off on me and i ended up in never never land (but i saw the same thing before so that's nothing new). this still leaves how to tell the ai where the track is. by marking the edges of the track (so having two contours) should be good enough (i think) to give the ai an idea of where to go. would this make sense?

--alex--