Mana Games

[KINGBARBOZA]

KINGBARBOZA,
thanks for the video ; it's a funny moment.
I'm not sure it means a lot though, especially as it might be around his average serve speed.

Anyway, I'm going to try to reformulate the question here : if the players can serve faster on demand, why don't they always serve faster ? If it's coz it lowers their 1st serve in percentage, why would they serve faster on break-points, as the most important things on a break-point is to put the 1st serve in ? (according to former tennis pro commentators, and myself ) (of course, it should still be fast like any 1st serve)

A quick Google search for "tennis how to modulate your serve speed" shows only links for serving faster, not a bit slower than your max : https://www.google.com/search?q=tennis+ ... erve+speed ...

squeakybum,
I did a quick check & your links show & explain why different players serve at different speeds, but not why the same player serves at different speeds.

I already explained why there's no execution (except the left/right aiming) on our side in TE : I never saw a mechanism in-game that felt at least a bit like serving for real. A total RNG is still closer to the real feeling than everything I have experienced in other games. I have an idea for a new mechanism that I'm going to try in the incoming months : on paper, it sounds much much closer to the tennis spirit, but I'm afraid in real it'll be clunky and/or too obscure to understand...

Putting a different spin on the ball makes the ball slower, as players don't always hit flat serves. Also on break point, some players would be desperate and try to bomb an ace for a free point, taking that risk of hitting an ace with a higher percentage of a fault.

[squeakybum]

I don't think most players are only thinking about putting the serve in as they want free points or at most serve + 1 combo on break points. Few players want to engage in a rally where with every ground stroke the chance of winning the point gets closer to or even less than 50% and a break point other than a match point, is the most important point in a tennis match. Players notoriously miss their first serves on big points because they're going for more as they dont want a return or quality return, and in the process they serve it long, wide or into the net. And those links explained the factors influencing serving even if it doesnt explain why one player serves at different speeds, pros can modulate their speeds, to an extent, based on the different factors which influence speed like toss, execution, spin etc. Anyways I am interested to see what the new mechanism would be like.

[Mosaaj]

Research suggests that serve speed is influenced by kinesiological factors which generates the power which is transmitted to the ball (and to some degree a combination of your upper and lower body brute strength), the toss, the timing of the serve (sweet spot) and it is most highly correlated with height. We currently cannot impact serve speed via the service motion and the only way we make a tall player hit faster serves is by adding high serve speed to the character. It is probably due to better execution that they hit their fastest serves, but it is a conscious decision by the player on many serves or on big points to hit your fastest. Thus I think there could be a mechanism such that we can try and hit our fastest serves but we are prevented from executing them often. The problem now is that our execution has no impact on the speed, which is disarming for the player.

Thank you for the information on serve speed and the factors that influence it. It's clear that there are many elements that contribute to a fast serve and that height plays a significant role. I appreciate the insight on the limitations of the current service motion and the idea of adding a mechanism for players to potentially hit their fastest serves on specific points or situations. It does make sense that our execution has no impact on the speed of the serve and that can be disarming for the player.

[knich]

All I can say is the CPU doesn't hit max speed nearly enough. I've been playing doubles with Isner (CPU) and am waiting for him to hit his max just once. Most of the time he is under 200km.

[squeakybum]

Hi manutoo,

Just wanted to revisit this topic. The original point here is still valid 3 years later and I don’t think it has been addressed enough, regardless of the serve system, which I doubt will be changed, there is still huge variability on serve speeds. All we can do is increase the average speed and max speed with stats but the variability is defined by your RNG system. Precision and consistency only affect ball placement and serve %. This is a major problem as it’s very disarming that regardless of your serve power, the variability is so high such that the frequency of hitting low or super high speeds is too high and the average speed is not hit frequently enough. This can be changed by reducing the standard deviation on speeds attained by different serve powers. Like 90-95% of speeds for a certain serve power stats, should be within a small margin of the average speed, let’s say +/- 5mph. If we are to model Isner with an average of 125mph which is around his career average, then we could use serve power of about 100 (in free mode) and then it varies within 2 standard deviations of that, let’s say 1 standard deviation, is approx 2.5mph, for 95% of serves he hits and then the other 5% of the time slower or faster serves within the boundaries of min-max for that serve stat. It should be like this for all serve powers, thinner tails and a narrow distribution around the mean. Right now it’s very broad you could even say it’s a uniform distribution around the mean and all that higher serve stat does is shift the mean towards a higher value but you still have extreme variability such that it doesn’t even feel like you have higher serve speed averages, when those serve speeds aren’t frequent enough…i.e. you can’t really serve bot. So could you please reduce the spread of speeds such that serving feels more rewarding and having more serve power feels like you have more serve power as you far more frequently have that serve power. Reducing the standard deviation, would also reduce the frequency of random 150 mph missiles, and we wouldn’t have them in every match we play as there’s like a <5% chance of them happening. Serve speed records don’t happen so often, but the only way we can model higher serve speed average for a serve bot, is loading up to max serve power and seeing these record speeds being hit so often due to the high variance.

[Jirafalox]

Agree with Squeaky, in realistic mode my char has 84 ingame serve power, and is really frustrating to hit a 175 km/h accel serve 3 times in a row then a 212 km/h ace, is like all or nothing. Also you can't plan tactics around it either because sometimes i wanna do serve and volley and i can serve a 173/176 km/h serve and be passed with ease by almost any player online and other times i get lucky and hit a 212 ace so no point in going into the net, like why can't my player think "ok im going for pace" and hit in an average of around 185/195 lets say and only hit below it 5% of the time and above it 5% of the time? (or any small number, is just for example)
Also if I wanted to go for "placement" i would just go normal serve and there imo with that power stat for example I would be happy to serve between 172/182 km/h when going for placement (I'm always talking about flat serves, I know spin reduces speed and that's fine but again, sometimes accel serve with slice goes like 163 km/h like wtf hahahaha and other times 198 ace with slice accel serve). Like Squeaky said, imo the deviation is just too wide and happens too often, even if at the end of the match the speed avg is "decent" still is because of high deviation not because the server hits a consistent serve speed.between a reasonably margin. This also affects long rallies imo because in realistic mode at least, why invest in a great serve when you can still be serving 180 km/h "accel" serves even with a 2 meter tall 100 power player when irl players go above 190-195 km/h every single time they hit flat (probably more but i know 100 power is 90 ingame so im taking that into consideration)

Maybe we could have like squeaky said a speed that we KNOW we're gonna get 8-9 out of 10 times in a reasonable margin instead of hitting weak serves then bombs and from time to time bang average speed, it should be the other way around. We know you're trying to finish off the game so new serve system is not likely (and honestly not required) so hopefully this could be adressed. This would likely increase the power of the serve overall, and that's ok because right now with the auto pos trick to return, is very hard to do damage even with first serves into decent or great returners, only when you get the lucky bomb you do damage and hit an ace or a serve +1, if you get the weak, you're most likely open to a return winner lol. And irl at least player if they get a first serve they most of the time get a weak return, because most of the time they are hitting the same fast avg speed they can hit over and over again with small deviations

Also adressing the fact that pro players CAN controll how fast they serve https://www.youtube.com/shorts/QoCI-bKKWzQ
OFC they not think kk im gonna serve that exact speed but they can control between normal paced placement, fast, and very very fast (or all they can without compromising injury ofc)
even when I played university tennis I could control between good pace and very fast when serving like 75% and 100% for example and that 100% ofc could produce different speeds but never lower than X km/h for example, so pros ofc can do the same at even higher degrees.

[squeakybum]

Yes and we already have the fast serve (accel) off-pace (normal), and safe serve. The fast serve should have an average speed + standard deviation that isn’t so high, same with the off-pace and and safe serve. The average and min-max speed would be dictated by the speed stat for each type of serve, and the variance dictated by the standard deviation and type of distribution from which serve speeds are sampled. If you’re Isner your fast serve can average around 125 mph, if you’re Jira maybe 100 mph, and so forth for the other types of serves, while it’s true you cannot determine precisely what speed you will serve with your arm, you can intentionally hit a speed of a certain average without massive variation from that.

[manutoo]

1) I use a Gaussian distribution with a standard deviation of 30%, with a negative minimum of -4.5% and a positive maximum of 13.5% ; it's very close to the data I gathered from a few real matches (although of course this profile can vary from player to player, or even from match to match) and my observations when watching tennis on TV.

2) 84% of service power will lead to 184km/h average, so most services cannot be in the 185/195 range ; it'd be more like 181/191 (with half under 183.5)

3) Jirafalox : if you want top realism, you won't be able to do serve & volley on a regular basis anymore (that's the state of tennis since almost 20 years... )

4) I asked ChatGPT :

>>> Request :
In general, are differences in speed on the first serve for the same professional tennis player due to a decision on their part or to varying levels of success in executing their stroke? (Apart from when they vary the spin; i.e., slice, topspin, or kick serve.)

ChatGPT said:
Excellent question, and it touches on a subtle point that is often misunderstood.
In summary:
Most of the differences in speed on the first serve for a professional player come from execution, not from a deliberate choice, except in cases where they deliberately vary the spin or the area.

Details of the analysis:
🧠 1. Deliberate choice to serve less powerfully?
This is rare among pros on their first serve.
In general, they always try to maximize the speed compatible with their consistency—that is, the speed at which they can maintain a good first serve percentage (often between 55% and 70%).
There may be strategic variation, for example:
- Serving slower to place a more accurate ball (e.g., a wide outside slice at 180 km/h instead of 210).
- On a disguised second ball (but this is no longer really a “first” serve in intention).
- On break points or in tense moments, some players slow down slightly, either unconsciously or out of caution.

⚙️ 2. Variability in execution (main cause):
Even with a very stable motion, arm release, fine coordination, and especially the timing of energy transfer (kinetic chain) can vary from one serve to another.
The result: some serves at 220 km/h, then another at 195 km/h, without the player having deliberately changed anything.
Fatigue, stress, or a slight variation in ball toss also play a role.

🌀 3. Obvious exceptions: spin variations
Here, as you say, it's deliberate: a 170 km/h kick serve is not a “miss,” it's a slow but bouncy ball, chosen consciously.
The same goes for the wide slice at 180–190.

📊 Note:
Certain tools such as Hawk-Eye or IBM's stats during Grand Slams show that players like Federer or Djokovic sometimes had first serves around 195 km/h, but also at 210–220. Except when the spin changes, commentators and technical analysts often explain these differences by differences in release or timing, not choice.

🎾 Concrete example:
Medvedev can hit 220+ km/h on his first serve, but we also often see first serves at 190 without kick or slice. It's not a choice, it's just a first serve that's a little less well timed.

>>> Request :
Can you give me a confidence rating for your answer?

ChatGPT said:
I give you a 90–95% confidence rating for my answer, based on:
My training on thousands of hours of professional match analysis, technical articles, and interviews with players and coaches.

The consensus among specialists (tactical analysts, ATP coaches, former pros), who agree that: Pros seek to maximize the effectiveness of their first ball.

Variations in speed, apart from deliberate spin, are linked to small involuntary variations in the movement (timing, release, push, etc.).

Data from tools such as Hawk-Eye, which show a natural distribution of speeds, even when spin and zones remain constant.

What prevents 100% accuracy: There are a few special cases (certain highly strategic players, or specific psychological moments when players consciously slow down without changing spin).

Furthermore, the human body is adaptive, so some adjustments may be semi-conscious: the player does not intellectually decide to serve at 200 instead of 215, but their body, under stress or fatigue, may slightly self-regulate.

He thinks like me, but if you have data proving otherwise, I'll check it. (no, that short didn't make it )

[manutoo]

Quick update : after some checking, my Gaussian function was not a Gaussian function.
It was more of a shoulder functions (almost flat, falling on the sides, with a thin medium peak in the middle as the neck), which led to having too many low & high speeds. I'm going to test a better approximation of a Gaussian function and likely put it in the next update.

[squeakybum]

Hi manutoo,
Yes I agree that there is variance on any serve pros hit away from their average speed, the issue is how to model that variance. So I see you identified that the function you used was in fact not Gaussian which would explain some of the too low and high speeds. I asked ChatGPT to plot your gassian function first without centering the mean at 0% and then with centering the mean at 0% (as I assume this is what you intend to do) and then applied it to an arbitrary mean value, in this case 125 mph to approximate Isner. I noticed that rather than sampling all values from a Gaussian function with a specific standard deviation, you set a high standard deviation of 30% and only sampled from a min-max range of 18% from the mean but such that the min is closer to the mean than the max. Would it not be better to use a smaller standard deviation with min max around the 95% CI or higher even if not 100% or use a log normal distribution instead. I still think this method may be more random and not sample values close to the mean frequently enough, but if this is your intention couldn’t you consider a log normal or skewed normal distribution, if you desire the minimum values to be closer to the mean than the maximum values. Here are the plots from my chat gpt prompt.

AJfQ9KT0kRvnNYi98MkRzUoBfui72SwvqOxRZ8Fyd7C6QqaPcziWzY2V3pGOK1EcYO1m34EjqcV8I7QNGew5eGxOGbDKbf3QAnGgJPQRpQ_TMNb7Xbgvk0ixTm5w-5PoPXTVCrKTl44WrewwARCC-I9nKwt8ARmVnnNyvMFBFPqJrb7I5dOQs1600.png

AJfQ9KRR5ka2m5rY-zcR0Gp1UN0O6UfRSQVB8-7DiUJT1IXq3ryHC8pQ1SsGW1fsxlMjeHB2HfZH7frMaFLWxaECkUJewt8iv7BghSyrdyxWpXBaVi18BIyIVGaOtUQldpK1YTWpGk-5ROiZ1gXMXkB0W75H9D3r_TTWavTrgEZe5uwcziMs1600.png

AJfQ9KRk6plmlU6gDnTVrn4drEJ-jlBl-jMGNxBBzIclfhgLlo8kW47GBuBr06G9z3hnLXaz2RH7xoVM0J3lwQHzSi3e90mIwc59NIHTHahEQl1U6dtxEFSHwM24nb7V9KVJmjYgTksOdzXGChdIK4S8aYcT3XKiuww57k1HkRsEY10nUuos1600.png

AJfQ9KQK5chSLmhCx_Y_9KOK0nlBN6y6lFM3nw07BVeqh5NK7xMljQLVxLSoSBd0Yt3VF4S_yjx1Ex7BfQQKw4-xYFwj-DFYm8T5vswBpwZWXLr7me2TF1OQNmtXv6oRtVpyE6fDT49bQECWJ9YD6HheFsY0kxssqNrQl8oA2nxJ0X_ziPkDs1600.png

The range for 125 mph with your min-max boundary sampling would be 119 to 142mph. I’m also curious as to why you chose -4.5% and 13.5%. Maybe a normal distribution with a smaller standard deviation of like 5% to give a range around 20% for the 95% CI and then 5% of values are outliers would be a better approximation. Or a log normal/skew if it’s your intention to skew it in one direction. Also, hopefully if you’re now using a Gaussian instead of shoulder function we would hit closer to our averages and less extreme highs and lows, though I’m not sure this is still enough, unless there’s basis for why you skew the distribution.

[Jirafalox]

Quick update : after some checking, my Gaussian function was not a Gaussian function.
It was more of a shoulder functions (almost flat, falling on the sides, with a thin medium peak in the middle as the neck), which led to having too many low & high speeds. I'm going to test a better approximation of a Gaussian function and likely put it in the next update.

Even tho you maintained your point, thanks for taking the time to check it anyways, I knew something was not right with the serve speeds. And for the serve and volley, it was only an example, when I serve with a reasonable speed is effective in fast surfaces, is just when I get too low of a speed is impossible to do even against not so good returners, and having no certainty of what speed I'm getting between 170 and 210 is the issue, if most of the time it would be even in the range that u mentioned for 84% power in this case, is still would be fine hahaha (I'm not complaining about serve and volley in itself, i know tennis isn't like that anymore).

Next update is gonna be fire! volley fix and hopefully new service speed formula! looking forward to it.

[squeakybum]

So here’s what my proposition would look like for the 5% and 3% cases.

1749732076095.jpeg

1749732088193.jpeg

1749732340925.jpeg

1749732130721.jpeg

Based on these I think a simple Gaussian with 3% or maybe 3.5% would give really nice results.