Right, I did see the crash once. I need a way to reliably reproduce the crash.
EDIT:
Trying a lot, I never saw it again. In my case it was an actual crash, but not in LFS.exe and I didn't have LFS in the debugger, so no info yet. I clicked to enter game from the entry screen and got a crash, but only once.
UnknownMaster21: You called what you saw a "softlock" but what do you actually mean by that?
It simply means "LFS is not responding". Of course, it could be rather a hard crash and not a soft one.
I got another one, which had exact same impact. INVALIDKA
EDIT: This happens when AI tried to drive this vehicle.
EDIT: I do wonder if the issues are related to "Configurations". But then again, I have no idea. As I am currently testing bunch of vehicles AI is attempting to drive. ( Have to start somewhere though for investigating this, in case it is related to LFS itself and not the mods )
Hey scawen! can you give us a bit more info on what was changed for the motorcycles? I know you said it affects lean angle and tyre forces but was wondering if we could get a tad more detailed
Thanks, I got the same crash with KART_DIRT and INVALIDKA. It looks as if this is not new in D44, so that's OK (I've made a note to sort it out). It's to do with vehicle settings the mod creator has used, that need to be bounded. LFS assumes the values are valid, then when AI try to generate a path, they end up getting a floating point error. It is not a problem for most mods which have the bump stops ("Max up") set correctly.
To be clear, is this something different from the "LFS is not responding" that you go before? I think it is different. Maybe "LFS is not responding" is a new problem. I guess that could be an infinite loop, resulting in a hang. I think maybe that could be from generating AI for a particular vehicle on a particular config... so if that is the case, I'd like to know which mod and track config caused that.
1) Insufficient lateral acceleration for given lean angle.
When testing for another bug (with a lot of bikes on track) I noticed that the mod "VELOCIPISTO" kept running wide. After a lot of investigation I discovered that any bike with relatively heavy wheels and a light bike would run wide. At first I thought it was a bug in the AI code, but much experimentation didn't solve the problem. I took the observed issue to an extreme with an extremely light bike and thin wheels with no rider. What was actually happening in that case was the bike could lean A LOT but the lateral acceleration did not match up. For example, by the laws of physics, if a bike has thin wheels, and rides in a circle at constant speed, then with 45 degree lean, the lateral acceleration should be 1G. However, this was not the case. For heavier bikes, the wheels are relatively light compared with bike+rider, so the anomaly is not that noticeable.
The actual physics flaw: It turned out that the weight of the wheels was applied at the contact patch, rather than the centre of the wheels. Through all the years, this had never been noticed. Actually, this is because the effect is so tiny for cars, as the wheels are vertical, it makes barely any difference if the weight is applied at the centre of the wheel or the contact patch. For this reason, I've left it the same for the cars, mainly to preserve the hotlaps. It is fixed in the development version for all vehicles, but in the public version it is in a special version of the physics update used only by karts and bikes.
2) Front wheels taking a wider line
I had kept noticing that the front of bikes took a significantly wider line than the rear wheels. It doesn't match up with my observations of real bikes. During my initial investigation of the lean problems noted above, I tried something which was to simply remove the effect of "camber thrust" and rely purely on slip angle to provide the lateral force. I was able to make this only affect the bike wheels, so that, again, the car physics are unaffected. In reality, camber thrust is a real phenomenon, so just removing it isn't really the answer. But the actual answer is to release the new physics that I have in development, and make sure that camber thrust is correctly simulated there. Apparently, in the current public LFS, it is excessive, or at least was excessive at large lean angles, leading to the strange slightly sideways motion of bikes around corners. Anyway, it looks a lot better now and I believe the bikes have a more stable lean angle when throttle or brake are applied mid-corner.
Yes, that is the case. This happens at least on AS4 and KY3 ( but KY3 not completely tested, but has happened on another vehicle, which is currently not available... mod I mean. )
I've tried improved bike physics and me as a superbikes enthusiast i found few issues. I like how it's more agile, pointy, idk how to describe it, less understeery? on corners. Surely does feel better in that aspect.
However i found out that bikes tend to be a lot more unstable now under braking. The death wobble that i was so trying to get rid off on my bikes and it sort of worked till D44.
Reaper X1 tubes and masses for example that were super stable under braking before.
I don't know whether that's because of removed camber thrust or because of wheel masses being at the center of the wheels now, but i don't really know how i could fix that now, if it's even fixable.
Turns out, the more tire pressure AI vehicles has, the more careful they tend to drive. I wonder how bikes will affect.
EDIT: Unless something else got updated in the meantime. At least AI takes corner more carefully, due less grip?
EDIT: Ahh "Braking prediction includes a safety margin to avoid late braking"... At very first AI did drive off the AS4 Corkscrew, but after editing a setup, that did not happen. According to earlier posts, it makes sense.
Interesting, thanks. I'll have a look at some point in the next few days. If you could tell me specific ways to check it out and compare the versions, that could be helpful. E.g. particular bikes on particular corners that show up the difference.
It's mostly noticable on any straight when hard braking from at least 280 km/h. Simpliest way to test it is to do it on Layout square. Reach 280 km/h with Reaper X1 or Chimera X1 and then brake hard to a stop.
No because it's not effective at high speeds and if you tap it while braking it will unsettle the bike even more as rear wheel becomes light so it locks up easier as well.
I think the new safety margin on braking is a bit too high now in D44, to the point that when racing with the XRR, if I'm on the inside line but slightly behind an AI, I can basically slam the brakes slightly later than the AI and get right alongside, and still have enough grip to complete the pass before exiting the turn without running wide. Maybe toning it down a bit could bring good balance between safe braking and performance? Or maybe have that margin reduced with higher AI skill?
Another point I saw mentioned a few times, which was introduced around D40-41 I believe (when AI stopped braking for no apparent reason), is the tendency for AI to bump into you, typically when you're on the outside line, both on turn entry and exit, where they can push you wide. I much prefer the current behavior to the old one, but I now tend to "protect myself" by basically pushing back against the AI car so I don't end up in the grass.
Somehow I still think AI braking simultaneously in the middle of the low-speed corner is pretty strange, and usually causes the second half of the pack to nearly stop for the first few corners at lap 1.
I tested the same way 180mph at a straight up stop causees crashable wobble
It definitely feels better but there is that now though.
Is it possible to setup a brake bais input?
Such as if we are on the brakes the pressure will be applied 100 percent to the front and lets say 50 percent to the rear? I think that would fix it as bike brakes work independently from eachother.
I have so far tested 562 different vehicles for AI in AS4. Depending on the mod and several situations, it is clear why tire physics must be finished. AI does not take any information about if vehicle is about to flip itself. I wonder, if that is the case in bikes which AI could go much faster if flipping is taken in practice?
EDIT: Now 598 vehicles. So far 3 vehicles can't make full proper lap in AS4, no matter what are the settings. But this is more like a mod issue itself and not LFS itself. 14 vehicles has several difficulties to make proper lap, but can still do decently.
( I am talking AI driving these with Level 5, PRO level only ) ( Furthermore, I am only using vehicles which SHOULD, supposedly do a lap by AI in this combo, as fast as possible )
As far as I know, one can't go faster by flipping the car over If the AI tries not to flip the car, they will finish the lap, but possibly not faster. I do think that it is a good idea for AI to be "aware" of the car being on all 4 tires all the time.
Thank you everyone for this very nice feedback and for Scawen who implements all these tiny but important improvements. Together, we'll make LFS awesome.
