Kennett's think tank - Lag and sync issues

[Valter Ostman]

Anthony and Peter, from my view your cars were a little twitchy in the beginning but stabilised after some laps. Never quite perfect though. I have seen your cars very stable and free from visible lag in many races.

 

[Nicolai Nicholson]

Thanks for descriptive video and comments guys

My thoughts:
1. The lag/sync issue was predictable in my eyes due to the high speed/movements and proximity of the cars. But it is good to have demonstration video of it to remind everybody.

2. The cut I see on the very start of the video is a big no-no. It happens from time to time with myself if I brake too late/turn in too late, but then I make sure I lose a little time on it so it manifests itself as the mistake it is on the time sheets.

 

[Nicolai Nicholson]

Another great Sunday practice stints with Valter and Peter THANKS GUYS ..( Nico and kurt were missing this time )..

I had very close fight with Peter for 12 laps, then when i tried to overtake him on the 130R while i made sure there was enough room to my right but again the usual thing happened ( the lag story ) .

I've been away this weekend, so I was not able to join the practice

[/quote]
I looked at your replay many times Anthony. My first impression was that it did not look like a typical lag crash, as both you and Peter were very smooth/predictable in terms of forward speed and there were several meters between your rear and Peter's front.. The way I understand the lag issue this should not have lead to contact. What do you think Kennett?

The second time I watched it I noticed that Peter seems 100% unaffected by the contact, is this typical? Does not both cars "make contact" even though it is lag-contact?

Then I think, that particular turn, for me I have to turn in at the right moment, and I am always close to running wide (using all grip availeble). You are taking it leaving lots of space for Peter (a more grip demanding curve), and your car is moving sideways a few times before your rear suddenly looses grip.

My point is, is it possible that what felt like contact was just your rear running out of grip? You know best as you were there, I just wanted to ask in case the possibillity did not cross your mind. And, that in my mind, the lag issue suddenly turned much more serious if this in fact was a lag issue.

EDIT: i might be mixing the names as the skins are alike and I do not know who is who, but my point should be easy to understand anyway I think? Btw, nice close racing before the incident happen

 

[Kennett Ylitalo]

I think it's pretty normal lag. The way the prediction works is that between packets, the car stays on it's current trajectory. Because we see the view from the inside car, it's trajectory is pointing straight to the car in front/outside. The collision happened only on the Peters end, Anthony doesn't receive any feedback since his computer says that there was no collision. It would be nice to see this from Peters view, i'll bet that Anthony will lag in his view enough make to a hit. So it was Peters prediction that decided it was a hit. Unfortunately, once it's a hit, the game can't turn back time to correct the false prediction nor can it simulate the forces that Anthony should receive for hitting..Those timeframes are gone forever..

Fortunately, this is the worst case scenario for the prediction system. Needs pretty exact terms for it to happen: short enough lag that the cars aren't made ghosts yet, the angle of the cars (trajectory is diagonal to the orientation of the car = max length of the vehicle..), 130R.. But then again, these happen quite often because we are very frequently on those situations.. Basically the car lags more front/back and less sideways but once we're in a very fast sweeping corners like 130R, it's more and more diagonal where prediction starts to crumble. It works fine on tight turns (less speed), straights (lots of speed but cars travel mostly parallel). WHen there elevation changes, the whole thing goes one dimension up and there can come even more strange looking crashes, cars can sink to ground or get airborne giving opportunity for cars to get on top of each other.

 

[Nicolai Nicholson]

But when do you assume contact happened? Before the sideways movement or at the moment of loss of grip? Or is the sideways movement just part of the lag issue (that from Peter's perspective the car does not move from side to side?).

Either way I agree that seeing this from Peter's perspective is very interesting and should confirm what happened.

 

[Kennett Ylitalo]

Yeah, i thought about this and my explanation is not good.. Maybe if i draw some pictures? The video does show the worst from the current system, you are much safer side-by-side than for ex right front close to left rear going +300km/h and turning. EDIT: I try to make some pictures

 

[Nicolai Nicholson]

Yeah, i thought about this and my explanation is not good.. Maybe if i draw some pictures? The video does show the worst from the current system, you are much safer side-by-side than for ex right front close to left rear going +300km/h and turning. EDIT: I try to make some pictures

I understand your explaination (at least first paragraph), it is the same way I understand it. But based on that I draw the opposite conclusion. But I am not saying I am right. What you say about contact only being experienced by one side sounds logic/correct (but it is still interesting with empirical data on this).

What I think is that if you make a sudden movement, it might take time before the other driver experiences that, thus making the positions of the car wrong for a short while. Therefore side by side is dangerous when making sudden/un-smooth sideways movement, while front/back is dangerous when there is a sudden or un-smooth change in speed (like braking).

I'll rephrase: If you go constantly in 300 km/h there is no problem with anticipating speed when packets are lost. They do not go constantly in 300 km/h, but the speed is relative stable, and there is a good distance between the cars. This lag incident is way beyond the safety margins I apply to the Aussies (although this is a case of 2x Aussie (doubble troubble))

EDIT: I do not mean to sound arrogant, I very well might understand this wrongly. (so I am not saying I know better than you )

 

[Kennett Ylitalo]

Lag in straight line:
- green is the vehicle
- red/orange is the lag
- arrow is trajectory

Lag in a corner

It's abit counter-intuitive for us humans, we can predict that the car doesn't suddenly lose all grip, gravity, wind resistance, yaw but most likely will follow the direction of the corner.. And we have to remember that there are two cars, not one, that has the exact same "lag-box" around it bouncing back and forth in different time but same space. One or the other gives away.

 

[Kennett Ylitalo]

Multiple cars in one corner. Even when the first car is closer to the second one, the situation is much safer. The more diagonal trajectory equals more problems.

I agree, the safety margins involving latency are getting ridiculous due to games inefficiency.. If it only took yaw into equation, we would have much closer racing...

 

[Nicolai Nicholson]

I am not claiming it works this way, it is how I have assumed/thought it worked.:

The direction and length of the arrows depends on the development of speed (speed curve) and steering wheel input leading up to the packet losses.

If your speed is constant and wheel kept at a constant angle, the prediction when loss of packet will be correct.

If your acceleration is constant, and/or you turn the wheel slowly and constantly, the prediction when loss of packet will be correct.

Iow it is not able to predict changes made while packets are being lost, it will predict you are continuing doing what you are doing.

 

[Kennett Ylitalo]

Like i said, if there were YAW in the prediction algorithm, the steering wheel angle would affect the outcome.. now it doesn't do anything. Trajectory is where the car would go if the whole universe is taken out, no road to have grip, no gravity etc., the car just goes straight.

The arrow/trajectory most likely is not straight in that corner for all cars since it moves all the time. It's simplified.

 

[Nicolai Nicholson]

I understand now I think(but yaw is a term/word I am not familar with, does it mean direction? ) And how is speed predicted? Does it lock it to the last known or "continue the development"?

 

[Kennett Ylitalo]

Here's a little bit more complex pics taking YAW in to consideration and each car has it's own trajectory. All cars are turning the exact same amount, the speed is the same.

YAW = the amount of energy that is changing the cars trajectory. it's a bit of a brainteaser to get it right. The force is actually torque and YAW rate (the speed the car is turning) is the "hp" or power of the turn. Those are not in any way the official explanation, just how i picture it.

Blue line is the cars path taking YAW in to equation and he cars here are equadistant to each other. Dotted black line is the trajectory.

Just picture how much more precise this would be if it would take yaw in but there's a reason why it's not there.. YAW can fluctuate wildly in each sample. It's YAW rate what we are after but that means some kind of root mean square sampling over a period of time, much more complex than a straight line.
EDIT: just saw this

I understand now I think(but yaw is a term/word I am not familar with, does it mean direction? ) And how is speed predicted? Does it lock it to the last known or "continue the development"?

Last sample.... i know, poor...it can have some factor that take the "development" in but after two or three missed packets in a row, there's not much to base the prediction to anymore..

 

[Nicolai Nicholson]

I've read your last post a couple of times, but are still unsure of the answer of this question; what happens with the speed? Does it lock it to the last known or "continue the development" (or something else)?

EDIT: If I understand your edit correct, the speed might follow development a brief moment before locking / becomming constant?
(something like. 296, 298, 300, 302, 304, 306. 308. 308, 308 etc. where each number represents a packet and grey is predictions.) Do I understand you correctly?

 

[Kennett Ylitalo]

Speed and direction does not change. So 300, 302, 304, 306, 306, 306, 306

With 50ms steady latency, we get ~20 packets of information per second. 300km/h that means 4,6m change in location per packet (83m/s divided by 20 packets/second)... So the prediction system isn't that bad at all, we can race closer than that in those circumstances. So i do think that there is "development" in equation when packets are coming in steady. It switches to different algorithm when the connection between two players and the server grows over certain amount that's unknown (unknown to me..). But that's only when we get these lag crashes too and the problem we are having.. When it's working, it works really well, we can't notice it at all.. But AFAIK it's only the last sample or last full frame if latency goes too great.

The biggest problem is not the frequency of the packets as much as lost packets or choppy connection.. Also statistically i think it's the safest option to go off track, the other solutions could make the car veer 90 degrees (incorrect yaw rate due to low sample count) in 300km/h or to stay still (do nothing..) both far more dangerous option. When the car show still, it's in "ghost" mode and no collision should be possible. i know, in practice the ghost will flicker on/off and can in some occasions "spawn" inside a non lagging car and other weirdeties

 

[Nicolai Nicholson]

Ok, so in a nut shell (and in layman terms), the car will continue in the same direction and speed as in the last packet until "connection is restored" and the car will reappear in it's real location?

 

[Kennett Ylitalo]

Ok, so in a nut shell (and in layman terms), the car will continue in the same direction and speed as in the last packet until "connection is restored" and the car will reappear in it's real location?

Correct. Once again page long explanation condensed in to one nice sentence

Just an afterthought, YAW rate is not being transmitted to telemetry but it would be nice if it was... Just picturing a nice bar in the screen showing the actual turning forces...

 

[Kennett Ylitalo]

Off-topic but what the hell. The difference in YAW and steering... That's me going thru the T1 and esses in Suzuka. Yaw rate is on top, it's just YAW samples over time. Middle is the steering wheel, bottom is the ground speed.

I'm overturning quite a bit, T1 i get the same YAW rate with 30% steering as 70%... but that's also the loaded tire losing grip etc. factors but truth is, i overturn in Suzuka.

 

[Peter Marshall]

Either way I agree that seeing this from Peter's perspective is very interesting and should confirm what happened.

Not that it is important, but I use the so called "real mirrors" [wing mirrors we call them], however, the video replay has decided for me to use the Virtual mirror. I believe that the wing mirrors are less useful here at Suzuka; perhaps because of the changes in track elevation that give this impression.

The incident was my fault [as is usual]. Whenever someone is on the inside blind spot it is always best for me to surrender position or it usually ends in tears. And I was not fast enough which is the norm for me.

 

[Kennett Ylitalo]

Whoah, that's a HUGE difference in those videos.. Much worse than i expected, like they are in different reality.