Scav Aim Is Fundamentally Flawed - Leg Meta Is Real || Escape from Tarkov Science

[@ACOGforlife]

As a scientist that is actually a near scientific level of analysis. You admit to some true statements about the sample size etc. VERY GOOD JOB, BSG needs people like You. Thank You.

[@alcatrazthatguyontheintern1465]

Gotta say thanks for your input. There's far too few people willing to admit their background in science, nor offer their input. Take pride in your work sir, and I wish you the best.

[@benjamingeris7890]

z Bogiem

[@jasonmurray4527]

i dont understand. you made a statistical video, and people are getting overly butt hurt that you used statistical vocabulary. This video wasn't bad at all, or is it the fact I sat through a business stats course that didnt use English, only statistical vernacular. If people don't want to listen to a video that obviously took you a shit ton of time to produce, and that you put a whole lot of thought into, then go watch any random video that will tell you what you want to hear without any fact behind it. I think you did a damn fine job with this video @Veritas keep it up!

[@dorHeadley]

I don't get it as well. I'm not even a native english speaker and was able to understand the entire video and used terms. Probably people should educate theirselves more.

[@OminousRain]

people are not annoyed over the information, they are annoyed by his use of too many words and their unnecessary use in this context

[@alcatrazthatguyontheintern1465]

@@dorHeadley I always appreciate people like you coming forward with your experiences. Alot of people have strong opinions regarding non-native language speakers obtaining information they need, but not enough people take the time to consider the opinions of people such as yourself. Regardless if you agreed with Vitas or not, it's a nice change of pace hearing someone be bold enough to come forward with their own experience

[@rally3927]

It's not the fact that he's using big words, it's that it's adding nothing to the video, in fact it's doing the opposite. It's frustrating listening to him take fucking forever to explain very basic concepts and findings because he's artificially inflating it with loquacious sesquipedalian rhetoric. See what I did there? This entire video could've been like 5 minutes and I'm 8 minutes in still waiting for him to shut the fuck up about meaningless discrepancies between sets of data

[@floridamancode_e2673]

@@dorHeadley same. He said like one or two terms that are related specifically to statistics I didn't understand but he specifically said that he wasn't accounting for those terms and asked people in the comments to figure it out if they knew stats.

[@dudeman7738]

Veritas: "Hypothesis" Everyone: *OmG sToP sAYIng BiG WoRDS HuRT BRAin*

[@sulkissulking]

using "big" words to describe something so simple doesn't make you intelligent; it makes you pretentious.

[@dudeman7738]

@@sulkissulking This is 2yrs old lol but hypothesis isn't a big word 😂 it's commonly used if you went to school it just means educated guess.

[@tankslayer3246]

Finally someone BSG will listen to has made a statement about this issue

[@tankslayer3246]

Hey Veritas I was wondering you could also cover the ranges of scavs with certain weapons. I’ve heard a lot of people say ,and experienced it myself, that they are getting shot from 50 or so meters and getting two tapped with shotguns like the MR ,MP ,and Saiga. I have noticed a drop in the amount of times a shotgun scav will shoot me from range though so I don’t know if they fixed it or not

[@TomatenDK]

Hahahha - just wait. The second he talks negative about this broken game, they will copystrike his azz :D

[@unit0007]

Yes, now all we need is for a high profile streamer to get banned by the broken anti-cheat. =)

[@tankslayer3246]

Victor Quinn hasn’t happened yet to my knowledge

[@derrickcoleman5481]

Honestly i dont think its an issue we are all terrible people and deserve to have our PMCs legs mutilated above all else LMFAO

[@zerohasbeendivided]

All these people upset at you using large words are super sad lmao. It sounds like you've written a research paper, and used it as a script. WHICH IS A GOOD THING.

[@OminousRain]

definitely sounds like a research paper but the teacher would probably give a better grade if the same information was displayed using less words

[@DOT107]

@@OminousRain less word? I mean, structure wise; the hook, body, and sinker are relatively the same topic repeated over and over. Hook< explanatory Detailed phrasing

[@tylerwatson7525]

ZerohasbeenDivided his big words don’t really work the way he uses them, so it just sounds like a dumb flex

[@sorrenzz3610]

@@tylerwatson7525 that is very true. It annoys me that it sounds like a flex.

[@Deadliner-svd]

Im 13 and not a English native speaker and I can understand what he is saying wtf is wrong with pple XD

[@jespermattsson816]

When using the aim offset scheme described at 1:50, there is a very simple mistake that would lead to the observed behavior. If the random distribution of the offset is such that the most probable value is anything other than zero, then the most likely spot to hit is on a ring around the aim point. At enough distance, that ring would then pass through the legs, but not any other part of the body. This would also explain the raiders on labs not having this issue - their random offset could be small enough that this ring still falls entirely in the torso. This is a very hard mistake to spot, since the logic looks very reasonable until you start plotting heat maps of the distributions.

[@tg5613]

Yeah I was thinking that same thing. If the limit of the offset from center of mass is consistent at all ranges then it can be seen that scavs would either miss or hit the legs. This could be proven by making the offset less at longer ranges.

[@youngburgerpatty]

wouldnt there also be a ring where headshots were also really common and somewhere in between always losing my legs and always losing my chest i would always just get one tapped to the face.

[@tg5613]

@@youngburgerpatty yes but you have to consider that the head is a much smaller target

[@jespermattsson816]

@@youngburgerpatty Consider that the legs are much longer than the head. It would be only a very small range of ranges where the head would be notably more likely. Also, you wouldn't notice it in the same way - instead it would be "Aw damn, one-tapped by the scav!", raid over.

[@jespermattsson816]

@@tg5613 Not quite, the offset would reasonably be an angle, not a distance. That would mean that the distance offset would be dependent on the range. This is what you want. However, you want the distribution to always give the highest chance to hit at the aim point, but be more and more flattened out the less the skill of the shooter.

[@subtlemonk]

2 whole minutes of scavs actively attempting to shoot legs, but cant......"theres no leg meta."

[@madzaisa]

Also: "Our SCAVs use actual ammo instead of magically taking it from their asses". Yeah....

[@mothorb]

to me it seemed more like they were aiming towards the stomach, one reason it might seem like they were aiming at the legs is that most video game don't put the first person camera where your eyes would be but rather the top of your head

[@mfenix88]

love your work mate and props to BSG for being willing to share the data, good to see players and Devs with a passion to improve an awesome game

[@T3HM4N244]

One thing that needs to be considered as well (I don't think I saw you mention it) is that it based on the 10:30 clip looks like whenever the AI sees even a part of your body, it still aims based on your entire player model vs aiming at what would be visible to it. This might be why people were able to side-step AI for a while and not be seen/hit when labs first came out and their solution was to just make it so that when any part of the body is seen, the targeting mode was activated, but still based off of the entire model. (this is just a blind guess/assumption though) Maybe someone can do some testing with side stepping/leaning to see if this is the case against AI with rifles with a friend to see what is being hit/aimed at.

[@partyboycs6086]

Yeah they should really only target what is visible to them and it doesn't seem to be that way

[@dazcookiez3714]

T3HM4N244 good point kind sir

[@fregyt]

If they target what is visible (which is what raiders do) you get headshot fuck tonnes

[@hunterscars01]

so basically in layman's terms, shoot for center of visible mass instead of at the center of the player model.

[@LPLimbos]

@@fregyt you could potentially adjust accuracy by percentage of full body in view or something of the sort

[@chris37477]

wow keeping in mind you are "just a player" of the game and not a Dev its insane how much work you put into the game.

[@coltonhertzog193]

Solid dood, just subbed!

[@sleeplessguardian]

video starts at 10:49 and ends at 13:24 xD joke assides, very informative video! keep the tarkov content goin

[@XD3blaze]

The quality of Veritas videos when it comes to game analysis is above even the likes of clicksphilipp

[@CaDzA818]

He needs to play some tarkov not just cs go :)

[@ascend2046]

And that rly says allot

[@ZunaZurugi]

Veritas is our clicksphilipp inc. the selfmade music :D

[@Reachermordacai]

Kliksphilip, fixed.

[@ElGreco910]

lol no,

[@TheBlankJoker]

When Military personal say Center Mass, we do mean the Center of the Chest. Its the largest target on a Human being making it the easiest to hit at varying ranges. So your first dot is indeed where we are aiming. So don't worry. We aren't trying to nuder everybody we engage with. :D But apparently these bots are.

[@nethengaming9]

blancjoker hey they just like to get cheeky😂

[@Warriorcat49]

I think the confusion comes from the fact that when we talk about aiming for the “center of mass”, we don’t mean the true CoM, but what is known as “upper center of mass”. It just gets shortened to center of mass, and the distinction is never made, leading to confusion.

[@VulpeculaJoy]

Center of larges concave area approximating a circle - aka CoM of torso

[@floridamancode_e2673]

Makes sense that a bunch of well armed hobos (meaning scavs) would do dick shots all the time though lol

[@specter2621]

Typically the phrase is "center of seen mass", because your targets will take cover, or be lying down or whatever else, and as you say in military terms it means aim at the big bit and has nothing to do with the actual 'mass'. In martial arts and physics terms, it is about weight distribution.

[@t3nply755]

This video (and other in depth analysis based ones on this channel) should be the STANDARD for how a creator should conduct themselves when referring to found issues in ANY product. The level of professionalism is absolute top notch. Excellent work.

[@tyranids2727]

Use the words vector and trigonometry, and Hayden's all around the world will lose their minds!!

[@albinospino3089]

Lol

[@mrjughead87]

I'm awwed in the information that you've taken the time to provide. When a fan and all around consumer takes this much time to help in the productions of a game that we all love it warms my heart on a level that I can't put into words. Also, the fact that you've been able to talk to the Dev lead Nikita and for him to take the time to hear you out and provide backend info about the game willingly again is awwing. As for right now, 0611 5/13/2019 you have a new sub and follower (me by the way). I hope to be in your streams and here on KZbin as much as I can. Have a great week Veritas.

[@Berachiel]

That part of just standing there with the legs behind cover were initially cool, then it turned to hilarious xD Just the way it dragged out was freaking great. Great video, once again :)

[@DaRealChrispy]

Amazing video. Great length, statistics and footage. I'm surprised by how much streaming, editing, video styles, and musical talent you have overall!

[@TheMeflon]

Nice work Veritas, hope Bsg find a solution fast based on your research.

[@garrettmccullough2249]

*Me reading this comment 8 months later* *Puts Surv12 Field Surgical Kit in Gamma Container*

[@elbarto8282]

13:36 Okay I have to say that surprised me a lot since I was expecting you to say it's OBVIOUS they were aiming at your STOMACH. I mean, the moment you got out of cover they shot you in the stomach, you had to heal your stomach 15 times, the part of the body that was always bleeding was the stomach, the stomach was hiding under cover and that's why you were not getting hit, etc. I agree with the rest of the video tho, thanks for the hard work!!

[@bastianstiefler3390]

As someone primarily sniping and primarily playing on woods I would like to add that Scavs seem to get deadlier the smaller u make urself as a target. Especially when laying down Scavs seem to have a way better chance of hitting long shots compared to standing. I will try to verify that in near future by testing, but it's something over hundreds of hours of gameplay that I consistently seemed to observe.

[@delta-a17]

Never dumb down these types of videos. If you want mass appeal you can start a "shorts" series or something where you talk like a 5th grader and list your findings but this level of granularity is awesome to those who are interested!

[@captainfalcon8424]

The hero we want AND the hero we need. I

[@4twstdmdl]

Two thoughts I've had are based on the AI targeting and programming: 1) Spatial reference, if the character model's point of reference starts at its feet instead of its center mass and the AI was told to aim at the character, then it would aim at their feet. I have this thought because there are so many times where the AI shoots at the ground around me. This could be tested by having the ability to look at AI vision and bullet traces in a debug mode such as the way ARMA 3 demoed bullet penetration. 2) Player stance, if the character model crouches or goes prone and then stands up then this information may not get updated in the server. So potentially you crouch or go prone, the server acknowledges you've changed your stance, then you stand up but the server doesn't get notified you changed back to standing, the AI then "sees" your character model most likely by ray tracing, then goes to shoot at your model based on some last known coordinate value but assumes it is in the crouch/prone position so it is shooting at you as though you are closer to the ground. This could be tested by performing the same tests you did before but making sure not to change your stance at all, if your numbers stay the same then it isn't this, if your numbers change, then it may not be directly this but would point to some kind of bug in the stance changes, whether its the server being updated, the AI being updated, or some type of position update being additive instead of replaced. Some follow-up: I do software development, I have done some hobby development in Unity which is what the game is programmed in. Point 1 doesn't feel the most likely because there is usually a lot of attention to detail in coordinate systems in FPS games, but it could be the result of a bug in the collision detection or ray tracing algorithm which causes the returned vector between the AI and the target to be looking at the x-z coordinates but not the y (assuming y is the vertical direction). For instance if the AI was at coordinates (x, y, z) = (1, 1, 1) and a target was at (10, 1, 15) then the change in coordinates would be (9, 0, 14). Assuming that the ground is at y=0 and the AI was told to turn to these coordinates instead of turn by the change in these coordinates then the AI would face roughly the correct x-z direction but would be facing at the ground. Point 2 feels more likely between my two points as it seems like a simpler edge case to overlook, and because we've already seen AI not respond correctly to changes in the player's stance, such as the side-step door peeking that was exploited after labs came out. Also because in my personal experience I have had a lot of times where the AI will shoot the ground as though I was prone, and I believe would have hit me if I was prone (15:02 shows this). It would also make that in-game clip (10:20) make sense because he was doing a lot of stance changes, so the AI's vision detection is seeing his character model and seeing a stale stance (such as still prone) so when it goes to shoot it knows the direction to shoot and shoots with a stance offset of prone.

[@TarkySUSorCheetahs]

Chris Logan I like where you started, which is why I lay down more so to shoot scavs in this update than previously. It is 100% accurate that if you lay down the mechanics of the AI is to shoot low, stomach and below, and the bullets will always hit the ground 5 or 6 times and seldom they hit you and that is long enough and I then shoot them; I too have been testing theories with the new AI tweaks on Interchange and my conclusion is that like almost everyone else, the scavs are aiming for your legs 95% of the time.

[@sterlingd1984]

Wow man love this serious degree of testing you're doing with the game mechanics. Helps us understand, the developers of EFT, and anyone interested in game development&design, how to better define variables of physics etc., keep up the good work man! Just subbed

[@riveraluciano]

I know I'm WAAAAY late to the game here, but what is most apparent from what I gather is the fact that center mass is indeed, center mass (a bit below the belly button, much like in martial arts as you pointed out, unlike what is considered center mass in shooting), this could match the general tendency to hit the legs. And yeah, it'd be great if the team of EFT could give you dev access, but you probably understand why they wouldn't. In any case, it's still amazing that they gave you that raw data and even more amazing that you went through the trouble of analyzing all of this, props to you.

[@NJOverclocked]

Yeah the data also doesn't account for the fact that most players don't peek Scavs with their entire body exposed. I realize that you covered this later on in the video, so well done

[@Andreloooooo]

Huge props for this work dude. Nikita should look into this with you.

[@chad4482]

You did fantastic research into this issue, and this needed to be explained in a professionalized manner. Disregard the negative comments, this was essential, you did great work here. If I may, Labs has a entirely different issue that you may be able to shed some light upon. The majority of runs I have done have been on labs including in 11.0 - 11.7. I have committed myself to a painful amount of hours on labs only totaling over 2k+ hours. I have rarely used armor, and played almost entirely solo. I have played offline, and online. The issue I will attempt to explain is related to armor accuracy and ammo used by AI. In my opinion. I can run labs without defensive items on my character and receive minimal damage to legs, but take most damage to stomach first then thorax. The defensive runs are primarily a stomach percentage depending on the helmet. If equipping a high defense helmet most of the damage was directed to the head when 7N39 was used by raiders. With equipping the "highcom ULACH IIIA" It seems to be a safer option to receiving almost zero damage in the head or face against raider. Again, in my opinion; This helmet seems to be the counter-meta for raiders regardless of the ammo. This may be due to the knowledge of the AI code recognizing the zero percentage of defense on the face. Equipping a helmet with a face shield I have noticed the raiders attempt to disorient the target, but without a face-shield the AI knows the defensive limits in the facial direction may be zero however choose not to aim directly at the face but instead still choose the center of mass. Depending on the ammo the AI uses, imo. It would be great to see the validity of this issue as we "ALL" know, end game needs work. Thank you @Varitas Cheers Spard :D

[@maksimrashkovskiy9187]

Leg Meta:Exists Everyone:Legs are unarmored Leg armor that could be put into the game:Am I a joke to you?

[@Pulowski]

overused meme format: exists you: it’s free real estate

[@maksimrashkovskiy9187]

@@Pulowski some randon dude who's life is that sad they go onto youtube comments to start a flame war: Exists You:It's free real estate.

[@Pulowski]

@@maksimrashkovskiy9187 that was cringe

[@maksimrashkovskiy9187]

@@Pulowski Says the one who's trying to start a flame war.

[@Pulowski]

@@maksimrashkovskiy9187 too late, you have already lost subscribers

[@Karatz84]

So, my observation... seems like the vector calculation from scavs when they are ads and not hip firing is off. Close range where they would ship fire they have a good aim point. But when ads the aimpoint is low.

[@suspectorange]

Or the point of aim is the belt line at distance. Which makes sense if you've ever kneeled behind a hill and had them shoot at you for a while, and makes sense from an eastern European conscription training perspective, although they should aim for exposed mass if the entire player isn't visible.

[@BallisticSollution]

This is some A grade work. Nicely done Veritas. I noticed last night that I was taking leg damage behind cover, cover that was about chest height. Now this could have been ricochet chance through "soft cover", but then I decided to run for harder cover, while running across the open I took more rounds in my legs blackening one and injuring the other. The other thing I noticed is that the scavs knew exactly where I was and when I was about to peak from cover to shoot at them. I think scavs are fundamentally broken, not just aim, but their ability to track you and notice you in low light conditions when no normal human could see you. It is frustrating, and I hope they overhaul scav AI significantly for future patches.

[@brassorchids4956]

as a biochem bs graduate, hearing all this data talk about a game I love to play is awesome

[@alcatrazthatguyontheintern1465]

As someone still in med school I've learned to appreciate it when someone else earns their degree. So good luck in your future fieldwork.

[@CommandoJenkins]

@@alcatrazthatguyontheintern1465 As a bullshitolgist, leet speak makes my vestigial tail tingle.

[@Th3EpitapH]

I remember sitting in rafters on factory night doing the sniper challenge and feeling genuinely 100% secure as long as my legs were covered. I just didn't get hit anywhere else. It's interesting to note that during that one clip, though, you were consistently shot not in the legs but the stomach. This could be explained by any of the reasons you put forth, but it's super interesting. I wonder if repeating the same thing at a greater distance would have given you leg shots instead. Either way, fantastic work as always my man.

[@xXDragonTribalXx]

So youve earned it Verts. Your new title is officially "Professor of Tarkov"

[@K7L3-93]

I recently unsubscribed from every single gaming youtube channel, which was around 50 individual channels because they lacked quality content and eventually steeped towards general shit-talk, drama and just outright immaturity, which no longer entertained me. However, the content you create is the exact opposite of that, and I applaud you, sir. You have my sub.

[@truedemoknight6784]

Perhaps cover causes the data to be skewed towards torso shots. When your legs are exposed, Scavs will hit your legs more often, but when they are behind cover, Scavs will hit your torso more often. This might cause the data to underreport leg shots, hence the inconsistency between data and player experience.

[@rgreenhagen]

One of the best analytical videos I have seen on KZbin, such an amazing content creator.

[@OneCritWunder]

What a great video! My goodness how much work went into this? Thank you so much for making this awesome video.

[@jonathanlebon9705]

I stopped the video at 6.30 to give my own opinion/analysis before hearing yours (so your opinion doesn't bias me in some way): My analysis of the month's worth of scav data from BSG tells me the legs get shot as much as the stomach if you consider both legs as one unit like the thorax..both leg hit percentages add up to 23% whic his as much as centre mass..distributed in 2. Next f you consider the fact the vast majority of players wear body armour, then your legs will automatically get hit MORE than the centre mass (unless the scav data only includes hits to a body part irrespective of armour, meaning a hit to the chest with body armour on, still counts as a hit)... Or does the scav data represent hits to the chest that causes damage directly to that body part? I will now watch the remainder of your video to see if our opinions differ. EDIT TO FOLLOW: Ok so, I watched the rest of the video and I will stick with my analysis but will add mkre to it: you get shot in the legs just as much as you do in the chest at low to mid ranges..at higher ranges their aim is off (as you pointed out) but this could have been a direct of the changes made by BSG ever since Scavs were changed from insta-headshotting, only thing is now they aim a little too low. CONCLUSION: Maybe BSG don't wanna raise the height because their data shows more headshots would be given? And I also don't know how feasible it is to get a scav to aim higher at longranges and aim lower at lower ranges, cuz having 1 level to aim/shoot at is both easier to code and avoids having additional low-level code to be processed, avoiding having your pc working harder to process the data, this also means it could totally make some guns/bullets fine but other guns with less bullet drop go too high (ie headshots)... I'd be happy to continue this discussion on the Tarkov Forums if you want? +1 for the vid...

[@epcfal]

Yeah. Given how eager they are to work on whatever thing peoples are complaining about on forums, i could very well imagine some dev "fixing" scavs headshots by making them aim lower at certain distances.

[@Urbanmember1]

what I really really really really like about this video is that you give the answer to the question right there in the title

[@icebite9888]

Thanks for donating so much of your time to help figure out the cause that has led to the demise of many victims of the leg meta.

[@trinston13]

All you have to do is play one raid to understand that scav leg meta isn't a myth. For Nikita to say it's not shows he doesn't play his own game and is extremely delusional. Thanks for the quality work Veritas. Hopefully this will open his eyes for once.

[@johnnymotorboat8824]

Thankfully now I get head/eyed from across the map. Perfect. Lol

[@jkoni]

I have some gamedev experience and I worked on some of these systems for my own projects. I think that the aiming rotation is calculated by 2 Vectors. One from the Camera of the Player/ Scav and the other Vector is the target point which the Scav is focussing on. Then you calculate the angle based on these vectors which then is applied to the animation system and handles the rotation of aiming. As Tarkov probably uses True First Person animations( First person animation is 100% the same as third person), this system provides accurate shots even though the animation leads the aim. I guess that the Scavs aim on the body depending on distance. Therefore, they focus on the selected bone of the character 3d model. I had situations where I walked up to less than half a meter to Scav and he aimed up high to my head. When their target is further away they then target one of the spine bones and the last point ist simply the "Center of Mass" which is the character center point also named "parent bone" ( which is superior to all other bones of the character). Of course there is an offset which can be driven calculated or is lead by animation and recoil.

[@jnarai]

Great video and explanations! The first thing that came to mind as soon as you separated the statistics by range was that possibly something in the (standard) scav aim logic is accounting for bullet drop, and the AI is trying to counter. Only it's inverted and increasing exponentially at too great a rate. Standard aiming of course is aim slightly higher at greater range to target, aim lower (lowest point being centre of mass) when close. To me anyway, it looks like it's doing it in reverse and like I'd mentioned - over compensating for the range. If the something in the aim logic were accounting for bullet drop, assumingly caused by gravity in BSG's accurate ballistics models if a (-) were misplaced or missing, that would invert the aim model for what we've seen. Scavs aim high when close (15m). Additionally, if the aim logic is accounting for gravity, if that value was increasing exponentially to a maximum (ie -9.8 m/s cubed vs. squared for example) that'd also line up with the impact points at different ranges. The aimpoint and trig ideas you had would also line up, hopefully BSG can sort it out though, whatever it is.

[@solemgameinsights]

Veritas. You are a gem of the community.

[@bodaciouschad]

Randomized vectors in 3d space allow for a HIGH frequency of non-factor vectors in the scope of an orthoganal reference. Simply put, if the aim adjustment vector is truely a randomized vector, the vast majority of directions the scav might offset it's aim by are not orthoganal to the line of sight between them and their target. This means that the actual offset that is expirienced, which is purely the displacement of the shot from the center of mass in reference to the axis of line of sight, is going to be cos(theta)(intended displacement) this means that it will average to be about half of the intended innaccuracy. Shots have an equally high chance of going over your head as hitting your legs.

[@ecklife]

Damn the amount of work that went into this video is crazy.

[@TheAngelsKill]

We dont deserve you Veritas, thanks for spending the time to compile all of this and everything else you do. Hopefully it will produce real change in Tarkov.

[@marcosgmunoz1]

holy shit man i absolutely love your videos. Did you work in data science or business intelligence or something like that? Where'd you get all that experience? applying scientific methods to literally anything you like is always so fun, I love seeing it done with games like this

[@SlimyTaco]

Well you learn a lot when you become a software engineer

[@VeritasGames]

Thanks dude! I'm a software engineer and part of what I've been responsible for at a few different companies was architecting analytics infrastructure and designing what data to collect. Having an intimate knowledge of the features and the data that's collected gives you a different perspective than people who are just looking at raw data. It's always a challenge to try to collect and share this data without skewing it or misrepresenting it - which is why this stuff immediately jumped out to me as interesting!

[@Ciraaxx]

Nikita: Why would I code the scavs to shoot peoples legs? Nikita: Yea I remember coding scavs to shoot peoples legs.

[@MrCrazeDawg]

Veritas, thanks alot for taking this time to analyse, let us hope Nikita will be able to fix the issues in the nearest future. #teamsoon

[@Just_A_Gun_Guy]

Makes sense, another good variable is that your body has more armor than your legs meaning you may percieve being shot more in the legs due to the facts that maybe out if 5 shots, 3 may hit your chest but not be that bad but 2 of those in the legs may black them. I'm curious if that also adds to the perceived targeting of the legs.

[@hijackdallas6052]

16:40 You can use the euro pallet as reference. They measure is 1.20m x 0.80m x 0.144m. The stairs are roughly 1-1.1m

[@Prmni]

Damn, awesome work and that was really cool of Nikita to provide that data!

[@archite001]

Great work. It's clear you put a ton of thought and effort into your videos. Honestly don't know how you do it man! Keep it up.

[@ewuf]

something that came to my mind is that there overall accuracy is based on the percentage they can actually see you, so if they see only 20% of your body their overall acccuracy will decrease by 20% INSTEAD of aiming for the parts of your body they CAN see.

[@rbbartos]

i have a hypothesis. the scavs may likely have a static crosshair placement. let me explain myself. a scav is given a vector to his targets COM (center of mass, whatever that may be in the game) and from that he calculates a default point of aim (presumably lower than the COM), disregarding the distance to his target. now, from the geometry of the situation it is obvious that the further you go, the "higher" your legs appear to the scav and the closer they are to his point of aim. just a thought.

[@TheFerwell]

This got my inner nerd tingling, thank you Veritas!

[@mattdayman9632]

Fair play for taking the time to put this together I would say it took a shit load of time good job.

[@JMPERager]

So, hide the stomach? Gotcha. I will become the Bane of Scavs!

[@themidnightanimal1555]

Armored Girdles incoming!

[@zeejenkins]

As a game developer, I think your hypothesis is spot on.

[@michalklucz6907]

again.. I love how detailed your videos are.

[@KugelXFanger]

Hey Veritas, one potential mistake is (and me and my buddy found out by accident): that the acctual camera of the player is more or less in the neck/top of the waist. They did this so the first person model of guns would line up correctly. I do not know how this would translate to scav AI (or even of it translates) but if true then their aim would be even more fucked up.

[@gemini_1085]

As a fellow engineer, I believe your hypothesis is correct Veritas. I believe the vector of aim is incorrect, and is being passed along as an incorrect vector. Excellent science mt friend, keep up the good work :)

[@luanmaia]

Man, thats amazing. Thank you very much for putting this much time in this. Great information and i think you may be right, actually. I would love to see this bug fixed.

[@lukatheslav]

Conclusion: If engaging scavs that are 15 meters or further away, try hide your legs behind cover or peak behind a wall or something.

[@tipsybutcher8941]

Fantastic video man! Really good stuff! I sure hope BSG gives you more to work with. That would be a very good thing for the game. Keep up the good work V

[@error-un3fo]

That’s dedication right there, loading times must of been a killer. A mod that gave you invincibility (for offline of course) would have been a life saver 👍

[@TrashPanda338]

Funny little add on, if i'm not mistaken unlike the american manual of arms which teaches to shoot "center mass" in the chest, the russian manual of arms teaches to shoot at the navel or belt line. This is to compensate for different battlefield zero's for different types of weapons. Crazy in depth video, thanks.

[@omarcintron1589]

Aye just coming to say they fixed it cuz of u thx my dude.

[@loungegod]

Very cool, thank you Veritas

[@SushiDushi500]

Great information. Usefull as always veritas keep it up

[@TheMojo354]

Veritas did his homework good job ma dude, Love this stuff

[@literallyjohn3046]

Another video helping push you towards full time KZbin. :) Edit: It's insane how much work you do in free time for these videos.

[@BlueFreeZeYuT]

What I don't understand is when you playing factory and if your on the big sky bride looking thru one of the slits in the metal wall and a scav starts firing at you from dead scav with a shot gun for example only your legs will take damage even tho your only exposing you head when peeking?? Just doesn't make sence only thing I can think would be the rounds penetrating the metal but wouldn't your chest and stomach take damage as well then?

[@8tracked]

Outstanding work. I also think it says a lot of Nikita and BSG to share the data for EFT research that may not come out in their favor. Keep up the good work sir.

[@Rexhunterj]

I'm glad they're willing to do so, it means there's a chance the many eyes of the community may spot the source/cause of an issue and that will assist the team in tracking down the specific root of the issue. In games like these, it can be as infuriatingly small as a single line of code that causes everything to go haywire.

[@8tracked]

Well said. It also, for me personally, removes pretty much all need for this game to come out !NOW! As long as they're receptive like this I know they're working to make the experience better and better, and I'm all about that.

[@combinemoosen3397]

Could be that it is a spread that leans downward so that scavs don’t insta headshot, and it doesn’t scale at distance so that it encompasses more leg area as you get farther.

[@Leepahhh]

Statistic significance in terms of the validity of data is generally set around 5% variance

[@1comoj]

Have you taken into account the zero distance of the scabs? If you zero with a 50 meter distance anything past that will hit low. So if you aim center off mass at 100 meters or more with a 50 meter zero the bullet will hit low.

[@DragonFoxStudios]

Probably this is caused by the scavs actively pulling the actor's location with a slight height multiplier which gets skewed over distance and ends up on leg height. The only thing I could think of that's causing it.

[@TopHatPenguin]

I feel like scavs are still insane after all of this time. I get one shot head/eye'd by scavs more than players half of the time and its driving me insane to the point where I just try to not bother with them

[@Bscoop75]

it could be origin of unity / physics body. I am pretty sure the "origin" of the player model is set to be on foot-level to actually place them on the map. it could be center of mass of physics model. but I am pretty sure, they did not set up an additional "center of mass" (which would be the chest) the enemy should concentrate his fire on. there is also an issue in Unity itself where animations can cause the center of a physics body can switch positions of not set up correctly. If I was BSG - I would take a closer look at the player model and investigate what part / config of that player model is actually the main point of targtting for the AI. as a test, I would add an additonal model (minicube, whatever) , put that right in the center of mass (aka the chest, the are where an amateur would try to hit and military personall is instructed to aim at) and make that the dedicatred target to shoot at for AI. then let rng and normal hit distribution take care of actual damage etc for the playermodel itself. although: the statement about the raiders makes me really wonder: did they built up raiders from the ground up ? surely seems to if their aiming and movement pattern are totally different as well. would be interesting to just compare raider shooting model (what are they aiming at) and scav aiming (what exactly are they aiming for) because the only constant is the player being shot at.. do they have inmplemented their test servers already ? guess not... great analysis mate. well done

[@greedkingcobalt200]

From a tactical point, aiming for the legs/groin makes sense since those are generally unarmored, unlike the chest or abdomen. both legs and groin also have _VERY_ large arteries running through them and even a non-lethal hit can (and likely will) cripple the target, preventing them from getting to cover nearly as well. Guess what we've all noticed Scavs doing? Popping our legs then rushing or flanking us when we can't run away. Could be a case like F.E.A.R. where the AI has a set of rules that it follows in unexpected and genius ways (F.E.A.R.'s AI actually had to be NERFED because they were so efficient at killing the play testers even on the easiest difficulty).

[@wesler218]

That's really cool he gave you that data

[@conberate]

I have a suspicion, based on your heatmaps for different ranges and description of the center-of-hitbox location, that scavs are programmed to aim for the center of the hitbox whenever the player is a certain distance away, likely to keep the random accuracy vectors centered on the target and possibly to reduce the number of random AI-inflicted headshots slightly. On the other hand, within the 13-15 meter range they would be programmed to center the aim at the upper torso because at that distance it becomes much more likely that the player is using cover, which would leave the scavs shooting into chest-high cover like in your video at ranges close enough to be painfully obvious.

[@Nick-fb4vv]

No one does these videos; and if they do, they're a copy. This is why I follow your channel.

[@haydenmcdonald4087]

Always love to see vids on video game science dissecting mechanics and "features" keep up the good work Veritas! Edit: Dad noticed me! Thanks dad!

[@SovietGrazz]

Perhaps scavs are compensating incorrectly for bullet drop: They're applying a curve that's a linear (mx + b) curve ... In the wrong direction?

[@macarebe]

Lol that part on the factory part of customs that the scavs shoot you for minutes without killing you would never happen on 2020s scavs. You would have been headshot on like their first clip

[@thenomad6924]

Out of curiosity, have you taken into account weapons sights? Contrary to popular belief, If you have any sort of sight attached to the top of any weapon, where you aim that red dot, is not where that bullet will go, even if the bullet we're to fly completely straight. There is an offset of, depending on the sight/weapon, an inch or more, with an even higher degree of variance at longer distance. This is why we zero our weapons/sights/scopes to specific distances and adjust, typically, on the fly by simply aiming above the target. In example, while simplifying everything so even the layman that doesn't own any weapons can understand. Say I have an AR-15 with an ACOG scope on it. Let's say that the reticle of the ACOG sits at EXACTLY 5 inches above the end of the muzzle. Being that these are both rigid systems, these lines will always run parallel to each other and never intersect at typical distances*¹ So therefore, when firing at distance, if you aim the reticle at the forehead, your round (If flying in a perfectly straight line) will impact 5 inches below the reticle, in the area of the mouth. While it varies wildly from person to person, the average zeroing distance is 25 meters. Meaning that at 25 meters, the bullet will impact where the reticle is exactly. I say all of this, out if curiosity to know whether the A.I. have a zero distance in their weapons, and if so, if they adjuat for distance in their aim, or if they will always keep the reticle at center-mass regardless of distance from target. While this may SEEM like common sense, you'd be surprised by how few people, even inside the world of firearm owners, know about this, and sense not even 2% of the population of the United States serves in the military, far less than you could imagine would know about this, even if this is a "Mil-Sim" *¹ This is assuming mid to close range; at distance in the real world you may have an overarching bullet trajectory where the apex of the bullet's arch crosses the straight line of the optics.

[@NoteNote447]

As a bit of a dumb theory going along with leg meta, and scavs targetting legs mainly at a distance, maybe somewhere in their code, it makes them target legs of players at a distance, thinking they're running.

[@PistolShrimpPimp]

I think if a spectate system was implemented it would also help see what's happening. Being able to spectate friends after we we die would be great as it is but being able to view Scavs in offline mode as a spectator in first person would give a lot of information, similar to Rainbow 6 Siege's Terrorist Hunt.

[@maluco132]

I came to a conclusion. The vector is calculated based on the NPC Eye level height, to the target center mass(chest). But then the vector is exported to the end of the weapon, without height/angle correction. The longer the distance the bigger the error. This could easily explain both data observed in yours test and samples, and also support nikita argument that it is working as intended. The error is in the execution, not in the theory/concept.

[@richsly99]

We're lucky to have you in tarkov community. Awsome videos.

[@tomstratton6120]

Have you thought about the muzzle velocity of the ammo used by scavs? Might be as simple as bad ammo not going where it should. Just a thought, since zeroing is affected by ammo. Not sure if it is the reason but maybe a low velocity ammo is throwing the aim of normal scavs, the raiders have better ammo and that's could be the reason for better accuracy.

[@nathangehman7018]

I think it would be interesting to have a test server for Tarkov where variables can be changed around for the sake of troubleshooting, but the changes are in no way changes that would ever be good for the actual gameplay; they would just be designed for getting data. For instance, if you wanted to double check that scavs are in fact aiming for center of mass, you could turn off bullet drop and give everyone random armor and shit and disable shooting. Their inventory on this server would be unrelated to the main server. So they would run in, stand still, get shot, and then we would see where scavs aim when bullet drop isn't a thing. Things like this.

[@Dubstepconcept]

Very well done man, you deserve A LOT more subs.