Wall Hitboxes
ฝัง
- เผยแพร่เมื่อ 8 มี.ค. 2016
- I explain how wall hitboxes work. I'm sure everyone's encountered their fair share of strange wall hitbox encounters, but fear no more - now you'll understand exactly what's going on and why! Thanks to Tyler Kehne and Peter Fedak for looking into the code for wall hitboxes and then explaining to me so that I could explain it to everyone else.
Here's a summary: Walls extend out in the cardinal direction that most closely matches their perpendicular direction. For 45 degree angled walls, they extend out in the X direction. Walls extend outward a distance that makes the perpendicular distance 50 units. They also extend inward a distance that makes the perpendicular distance 50. If Mario enters this parallelogram, then he is instantaneously moved outward perpendicularly to the edge of the parallelogram. - เกม
i was wondering why some walls in this game have such weird "invisible walls" around them for all those years. finaly it all makes sense!
nice!
Well, 7 years later Pannen made an even more informative video on invisible walls on his main channel. Should check it out
@@crylune i did watch the whole thing already.
@jan050375 my man 💪
"now that we're wall experts"
Yeah I'm a wallintoligist. I've been studying how walls work for 24 years
I'm sir wallington and this is my wall shop...
Walluigi is too
now 24 and 01
wall lore
I like how even the most basic, simplest thing in the game still makes your mind turn inside out.
But how do you turn a sphere inside out?
I remember that video it was so old but I still watched it all.
Peter Peterson
Do you mean to tell me we can turn a sphere inside out but not a circle?
+Teh Joosh Yes, Think of it like a monorail.
*a bunch of low drone-like sound effects*
Probably the most amazing wall video I've seen in at least two days.
tenchimuyo69 Lenny face
And now there's an even more amazing wall video out.
@@ianmathwiz7 Can you believe I have already seen it!?
This is surprisingly interesting.
I always wondered why the distance kept to some walls was so big, while that of others was practically nonexistant, and why you could sometimes walk closer then usual just before getting pushed out.
Great job on the illustrations, that made it very easy to understand.
Yay, now we are all wall-experts. ^.^ Did you prepare certificates?
A lot of these things sound like workarounds/small errors where the developers probably thought "nobody will ever notice or care". But they surely didn't expect a pannenkoek to exist.
+Lugmillord I wonder if SM74 now seems different now with wall knowledge
Chang Jia Yang
Not really. It's still the cheap hack it was back then xD
A wild Pannenkoek appeared!
Lugmillord Nobody expected the Dutch to make pancakes.
Things were done these ways usually due to a limitation in the hardware they had available at the time (things like the memory of the unit would limit how much information they can store in any scene at a time etc.)
Dude. I do not understand how you did this. I played Super Mario 64 years ago, got 94 stars, beat bowser, and then never really thought about it much after that.
Then I saw your video about killing the Mystery Goomba in my recommendations, and I was like, "ok, I have to know what that's about."
And now I'm learning all kinds of stuff about SM64 that I never even thought of before.
Your videos are hypnotically informative, and I love them.
So mario 64 walls are more complicated than real life walls
German Shepherd yea
*every video game wall is
Even though real life walls can be complicated too if you go into detail...
German Shepherd Not even close. Mario 64 walls don't have to worry about the atomic polarity difference between the human hand (and all its trace elements) and the surface of the wall (including its base material, composition of that base material, and everything attached to that base material by polar and other forces).
Touching things IRL is stupid complicated, and that's just for a basic understanding.
German Shepherd
Not really sometimes my hand clips into the wall
@@thatoneguy9582 i like how you make joke while everyone else is making logical statements
what the heck happened at 4:25 when mario went sideways?
Weird things can happen when you're crouch-sliding and crawling if there are walls nearby.
have we found any way to exploit these weird interactions? also it looks like it pushed you up, could something like that be what happened in the TTC upwarp? Have we experimented with this?
It's purely a graphical phenomenon. Mario's position is still on the ground just as normal
+pannenkeok2012
And it doesn’t happen in the mirror’s reflection in the mirror room for whatever reason.
See th-cam.com/video/KAZjj2BTpD0/w-d-xo.html
3:58 I'm a wall expert, mom!
Mom, get the camera!
What wall collage should I go to Mom?
Cheryl Della Santina
*wallage
*knawledge
Team Rocket do your intro next time
Question: Do the actual hitboxes (where Mario has to punch or kick to, say, break a cork block) correspond to the collision detection/bounding boxes you've described, or are they separate? I know in most modern games bounding and hit boxes are usually separate but given Mario 64's early release in the N64's lifecycle and that it's one of the first 3D games Nintendo ever worked on I'm not so sure.
Dang, this is such a good question gone unanswered.
Geometry detection and hit detection are most likely to different systems as most objects probably aren't solid. I don't haven't played this game much so this is just theory with no testing done
I think they were a lot closer to the visible mesh 'cause the N64 could only really render the necessary geometry for a level. later games added polygons to the visible meshes which would strain the CPU in collisions but mostly just increase the clunkyness of movement.
Also not sure what level design tools were used, may have been easier to just build levels in an ordinary 3d editor and write code to interpret the collisions than to write a custom tool.
I believe "If mario enters the 50px boundary while in x state, perform x action.
Not *one* of the first
But the first 3D game Nintendo launched.
I just want to say, your videos are amazing. I've linked them to a few of my friends who are like minded and decided to either go back to school for game development or started programming games from their general programming knowledge. And your videos do an excellent job explaining concepts that might not make a lot of sense to people intuitively, like hit detecting (collision), memory management (which isn't nearly as much of a big deal as it was on the n64, but still a fundamental you need to know), etc.
And plus, sm64 is a game they've all played so it's even easier to understand. I've taught them how to use the sm64 decomp repo to modify code and then run it, and this is a quick way of reminding them why they want to be a game dev if they ever feel overwhelmed.
As a programmer for games, i can't even work out why the hitboxes extend in that weird cardinal direction skew fashion. They already calculate the distance along the normal to extend it by, so its not even aligned to some easy grid or anything. Nor is it easier to test for collisions with, as the easiest to test are boxes aligned to the world axis (AAB, axis-aligned box) followed by a slightly tougher arbituary rotation of a cube (OOB, object-aligned box) followed by the tough arbitrary polygon shapes.
Also, as a bonus topic: the way OOBs do hit tests with a point normally is that you just rotate the entire world matrix by the inverse of the box rotation. Ie reverse the rotation so that it's an AAB again.
If you try to do a OOB hit test with any other polygon shape the same applies since it's easier to test against an AAB. However, if you're trying to do a hit test between an OOB and an AAB, rotating everything gives you... an AAB and an OOB! so no benefit at all
+Jewel Ots SM64 was coded on drugs
is it cuz Object-aligned Box is OAB?
+Jewel Ots Keep in mind, SM64 was one of the first games with a fully 3D world, and likely Nintendo's biggest attempt at it at the time. Mostly of the earlier 3D stuff, to my knowledge, was done by the western world and either had the 3D height done seperately from the rest of the physics ( see: Doom and the other idtech 1 gang ) or ... actually, I'm not sure how Build handles 3D height, but it's always looked janky to me.
Quake was released literally the day before SM64, so it's obviously impossible for it to be an influence.
I'm just wondering why they chose to represent Mario as a point instead of a sphere - sphere/plane collisions are way easier than points/axis-aligned-not-really-boxes.
My guess is they were just trying to save cycles. Checking distance to a _capped_ line takes a few multiplications, but their method lets you skip a lot of math with a basic "is pMario.y > topPoint.y or pMario.y < bottomPoint.y", and even more by just checking the distance to the _infinite_ line.
Yeah, and I'm just wondering why they chose to learn addition first - square root is way easier than addition.
My guess is they were just trying to save children's brains.
this is cool
this video seems to apply to OoT and MM as well, except the push distance is 15 units in those games.
How large is their character in units?
i wonder why the cork box's hitbox isn't aligned the way it should be, could it be because it's not a perfect square?
+ahbeef Probably. Maybe because it's...just one unit off
Damn, I didn't think it was that complicated. Interesting observation.
join us next episode where we talk about floor hitboxes.
Umneriko that legit happened
Umneriko it’s called “Walls, Floors, & Ceilings”
Could we theoretically use the slanted walls to warp through a thin wall?
I clicked on this by accident, but then I was already invested. What am I doing with my life right now, I have a paper to write.
6:18 "Inner Hibox"
My inner Hbox plays jigglypuff.
my inner hbox sleeps.
my inner box identifies as cork. it's not a choice.
typotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypotypo
I can't believe I'm watching an SM64 science video and enjoying it as much as I am.
tbh this is the only channel that I actually pay full attention to for educational purposes
I love these videos. Incredibly informative. I'm super interested in the little intricacies of design like this.
This is so super interesting, I've always wondered about the weird ways the objects pushes Mario around in this game. I'm not a very technical person, but you explain things very well, thank you for making these!
I think those snapping effects that occur when you enter a hitbox from the side could be very useful in speedruns. They could probably be used as small boosts to save a few frames.
Also, have you checked if there is no 1 unit wide gap between the hitboxes on the x axis of a 45 degree angled box? A single unit would allow you to walk straight through.
So much information, damn!
Wow, that's... awkward, but surprisingly simple for 3D. But seems like if you have a box at *almost* 45°, you'd have some funky angles. And what about slopes?
+Rena Kunisaki Slopes≠Walls
So only completely vertical surfaces are considered walls?
Perfectly vertical = walls
Facing upwards = floor
Facing downwards = ceiling
This is especially useful for triangles that rotate, such as the triangles that make up the red coin spinners in TTC. They are constantly switching between the 3 categories.
Huh. I'd have thought triangles usually stay as one type. I wonder if that could explain the upwarp? Some obscure corner case where a rotating object changes between wall/not-wall and a plane of its hitbox briefly becomes extremely long due to it being at 44.9999999° or such?
Except the spinners weren't rotating in that video since he had entered on 12
These videos are the kind of resource who's value doesn't become apparent until some SM64 speedrunner in 2044 uses this hitbox behavior in a strat
Wow. That was surprisingly interesting!
4:25
Mario defies the laws of physics.
This is quite fascinating. I wonder why they were set up this way.
Responding to the cardinal directions sounds somewhat reasonable, because the math can be computed much faster. But they compute the exact angle anyway to push Mario back, so why not just check it against that angle to begin with?
I've been playing through Super Mario Sunshine recently and I noticed some similar behaviors when colliding with walls and platforms. I assumed that Sunshine took the 2D Mario style of collisions (invisible lines on Mario's left, right, top and bottom) and adapted them to 3D somehow but this makes more sense.
At 0:51. That box on the left is one heck of an optic illusion.
Hence why I made the executive decision to show a less accurate by much simpler version as well for each diagram
the legend is back
These videos are the best! I love pretending to understand stuff!!!
as a game developer this is actually really interesting! I love reading about how older videogames handle problems like this
Was actually glued (snapped?) to this video. Incredibly interesting, somehow.
is there anything in mario 64 that's as straightforward as you'd expect haha
+Gronklems you can walk with the analog stick
+JopYT But not necessarily in the direction you are holding ;)
+JopYT but the tracking system.the inputs collisions,speed?
what have i done
I just remembered, Tyler made some slides about the mechanics behind walking: drive.google.com/file/d/0B06weeQA2uS9Vmw1NEFJS1E4WTg/view?ths=true
One of the best TH-camrs out there right now. :'3c
This seems like a VERY convoluted way of doing hit boxes. I can only assume this is because it was one of Nintendo's early 3D games.
As it's often been quoted that Starfox2 was one of the main inspirations for Mario64, I wonder what the hitboxes on that title look like..
Why is this so interesting its literally talking about wall hitboxes in mario 64 what has my life come to
I didn't know I wanted to know this, but now I know it, and I'm glad I do.
2:32 Goodness me. Is this an explanation for at least some of the "invisible walls" I encounter when I longjump?
Nope. But stay tuned for a video all about those invisible walls
i don't know why I'm watching this, but it's intriguing no less.
Love the explaining music
This fucking channel is deep like seriously it's just a wall
Well as this video explained, in Super Mario 64 the walls, *aren't actually walls*. They're areas which push Mario back, they just happen to look like walls.
It's more that walls are floating textures that have wall hitboxes around them, really.
6:16 Hiboxes are great, aren't they?
so is mario's position for wall collision a 2 byte short as well? I keep thinking that for mario to be "exactly" 6:30 at an intersection, it'd need to be in terms of integers (as opposed to floating point)
To encounter the snapping phenomenon alluded to at 6:30, Mario's position just needs to enter the triangle region where the two wall hitboxes overlap, not an exact point. So it's more lenient that just a single point. Also, I believe Tyler's told me that for wall collision Mario's float position is used, and it is not converted to a short first.
What is the pole in BitFS? Is that treated like an octagon you showed here?
Could you orientate yourself on the pole so that you have the maximum displacement possible? Even if it still doesn't allow you to reach the platform, it's still minimising the distance between Mario and the platform?
Jonathan Mace climbing a pole doesn’t use collision on the pole. It probably doesn’t even have collision.
So if you pick up an object like a cork box and rotate it around, does that mean that the hitboxes will transform to remain along the cardinal directions, creating a different topology of edges and lips? If so, can you think of any ways this could be taken advantage of during runs?
No. As seen in another video, objects picked up are not the true objects, but a representation of the object slot. the real objects remain where they were, but are invisible and intangible.
No. As seen in another video, objects picked up are not the true objects, but a representation of the object slot. the real objects remain where they were, but are invisible and intangible.
That's good info, but it sort of bypassed Leonhard Euler's question. What about rotating platforms then?
I don't know for sure, but I would guess that rotating platforms have their own, more computationally-expensive collision system.
Rotating platforms don't have walls.
Walls are weird in this game.
I guess, this early into the advent of 3D gaming, they were still trying to figure things out.
So what's with the weird hitbox on the star statue in the back of the castle?
And it's these gaps, misalignments and overlapping regions that produce the effect commonly known as "jank."
This is very interesting. Collision and hitboxes aren't really an issue with Source, at least not at this level of architectural detail, but this is very reminiscent of the visleaf system in Valve's (in)famous engine.
It's amazing seeing how programmers did stuff like this in the early days of 3D, when there was no tried-and-true standard way of doing things. Everyone was just fumbling around in the dark, building patchwork solutions, until middleware like Havok came along and offered a holistic, standardised method.
I wonder why they went for such a complex solution to something you'd expect to be so simple? Perhaps there were some limitations on the N64.
Nothing could be simpler.
This is one of the weirdest way of collision detection I've ever seen.
+Nick TwoThousand nahh
*****
since i am a dev, i know how its not hard at all
*****
videogames, apps and servers
hi my cringy version nice to meet you
LOVE the music
really interesting stuff :o
Thanks, now I know how to get inside your walls
Now I know why I hated the orange blocks so much.
... Why did I just watch a 7 minute video about the wall hitboxes in SM64? I haven't played this game in 10 years and I probably never will again
what about in WF? there are these pillars by the In The Blue star or whatever which seem to have displaced hitboxes
Huh, a good content creator, who doesn't speak or make meme/TTS videos.
*Amazing*
theories and guides about Mario 64? Mommy, where am I!
Actually, I'm pretty sure that Mario's environment collision is handled by a separate, usually identical though sometimes with minor alterations for inacessable areas (invisible walls) invisible 3D Mesh of the world or object (this was the initial solution to the 3D collision question), with normals with which to calculate direction, angle, precisely what a normal is. It's the normals that allow the collision equation to work variably, and also why it tends to fail on very acute angles - in these instances, normals would intersect, sometimes causing a collision miss, basically, a !=. It could also simply use the level geometry, since it never uses vertex colors but instead stores 24-bit normals, XXXXYYYYZZZZWW, instead of 24-bit RGBA per vertex colors. Normals are also used for the lighting equation, basically telling the light source the aforementioned info about the surface so it knows how to light it appropriately. A big problem with fan creations is inaccurate normals and/or color combiner commands leading to a lack of lighting or 1, 1, 1, 1 lighting. For an example of proper normals, take a look at basically any in-game object and notice how the light(s) diffuses across the surfaces uniformly.
However, if M64 does use boxes for collision, that'd be downright weird. Think about it - mario accurately portraying the illusion of interacting with complex, highly varied in terms of shape, scene polygons with purely squares? I'm not sure how well that would work in such a game... how you could create a hitbox for a world. Hitboxes (or spheres), work in things like fighting games, or shooters, where you just need an edge to edge approximation of the object with which to register hits.
Jeremy Johnson Afaik it uses this accurate to a 32 bit floating point. Also Mario can snap like 79 units up or 110 units down or something.
Jeremy Johnson no, the hitboxes are boxes that are inside the object itself.
Thats why Mario can sleep when he stands in front of the corner of a box!
are the hitboxes the same in a parallel universe? and can you do half an a press?
Well, for my game i am using the same testing method.. Using plane/sphere intersection.. For the corner problem i combine plane/sphere testing with sphere/edge testing.. Which will test a points distance against a line segment (a triangles edge) This way, your corners will turn into sphere segments which will repell you back, the flat walls stay as boxes.
I'm gonna play the panpipes whenever I need to go into detail.
For the cork boxes is the one side not at a 45º angle or is the hitbox just strange. If it were slightly angled, wouldn't that make the inner hitbox of the other wall actually extend past the edge of the outer hitbox ever so slightly?
You're assuming that the walls form a perfect square, which I said was a false assumption. The angles of the walls are slightly off from the ideal (and not all off in the same way), and for a wall whose ideal is 45 degrees, a small discrepancy will make the difference between an X projected hitbox and a Z projected hitbox.
+pannenkeok2012 So all of the walls are off, not just the one in the bottom right corner?
Some are exactly right, and some are a little off (in either direction). But the bottom right wall is the only one whose being off caused its hitbox to deviate from its ideal orientation. The moral of that example is that even when you try to apply the knowledge from this video to walls that you see, there's still some chance of error due to the walls being imperfectly placed
+pannenkeok2012 I guess I was just assuming that the rest of the walls were all perfectly angled. Thanks for the explanation. :D
4:25 he is on the pipe sideways. How does that work?
I think the hitbox extends 50 units as marios hitbox is 50 units from the actual position so that he wouldn't appear inside it?
Why'd Mario go sideways for a few frames at 4:25?
It was a prediction of Tony Hawk's Pro Skater 5
Whenever Mario crouches, slides, or does anything that allows him to be tilted left to right, forwards or backwards, a triangle (that isn't rendered) is used to calculate what side and amount Mario should be tilted. The reason why Mario went sideways is because of a glitch in determining where the triangle should be. Pannenkoek2012 has a video about it somewhere... Or maybe someone else made it, I can't say for sure.
Daniel Conley It was somebody else
How are square hitboxes different from something like a square object rotated 45 degrees? Could a box technically become a 45 degree rotated square if Mario moves into the correct position or moves it in the correct position somehow, or is it based on the box's position on the map?
It's based on the box's orientation relative to the map. Every map is laid out on a coordinate system.
+pannenkeok2012 ok, that makes sense. Like the box on BitS is right on the x and y axis perfectly
This video might be helpful: th-cam.com/video/3lADTcTkAvk/w-d-xo.html
6:13 Hiboxes, those are cool!
Very interesting! That geometric diagram implies that the cork box is a kite of some type.
Is it possible to end up too far inside any given box for the hitboxes of the walls to move Mario?
Yes. Anywhere on the left diagram, if Mario is in the center of the box where no wall hitboxes are, then he won't be affected
pannenkeok2012 Okay, cool! (Bear with me here, I have a few thoughts) Do floors and ceilings (i.e. the top and bottom of the box) have the same sort of hitbox as the walls?
Sort of. Floors are like rectangular prisms whose height is 79 downwards and they snap Mario to the top. Ceilings are like are like rectangular prisms whose height is 161 downwards and Mario can't voluntarily move into or above this area.
pannenkeok2012 Would it be possible then, for Mario to get in the middle of a box (or box-like object) and move to touch the box top's hitbox?
Yes, then Mario will snap up onto it. That's also how misalignments work: Mario gets under the top of the box without being pushed away by the walls. Hence, he gets to snap up onto the floor without beings pushed away by the walls.
Congraturation; this story is happy end. Challenge again? Go ahead dauntlessly! Make rapid progress!
Get out.
Wrong game.
Hmm... It seems that the diagonal walls extend their hitbox beyond 50 based on angle...
This suggests that the hitboxes were actually supposed to be at angles matching the visible faces but the devs didn't manage it for whatever reason, however they hadn't corrected the distance amount back to 50, so the game (incorrectly) calculates the distance by angle but places the hitbox aligned to axis.
What you think Pannen?
I'm just trying to figure out the "why?"
only panenkoek could make a video about walls and make it complicated.
I am skeptical that walls actually have hitboxes around them. Aren't they just detected by sphere and ray casts, giving you the *impression* there's a hitbox around them? Also this collision system very much looks like something that was only used for movable objects, like that crate. Am I wrong?
4:25 why does he go sideways for a couple of frames
Great video, but I don't see any justification for why the rotated hit boxes would be slanted. It must depend on the game or the engine.
what should i do with this information?
Why are the cork box's hitboxes different from the other boxes?
My best bet is that its considered as 44 or so degree wall. Like that 30 degree wall on 2:35
that's great and all but can you science how to cannonless please?
What was music for when you were explaining?
Mithel Celestia file select theme
Would I be right to guess that Super Mario 64 is its own engine?
If so, that explains why there are so many things to explain about the game...
So now we understand how the hitboxes of walls work, but what about objects, what are those hitboxes like?
I know that they're probably just a shape, such as a rectangle or square, but if they aren't, then how do the hitboxes of objects like Mario work? How about Bowser's submarine?
Please answer about the submarine. The sub is able to be cloned and it takes up a slot, so it is an object, or at the very least different than a normal wall.
Triangles are either walls, floors, or ceilings. A triangle whose face points upwards is a floor, a triangle whose face points downwards is a ceiling, and a triangle whose face points sideways is a wall. Consider the red coin spinners in TTC. The triangles on those objects are constantly rotating, and so they constantly switch between these 3 categories depending on which way their face is directed.
when its showing gameplay, which level/song is playing? thx
J4ckK file select theme
Is this just how they work in SM64, or how they work in every game/most games?
Do games still function like this to this day? Or have newer better tehcniques arisen as the technology improves?
Question slightly related to this, but what causes the "Invisible Wall" bug? e.g.
clips.twitch.tv/WanderingTentativeFrogKAPOW
clips.twitch.tv/GoodStrangeBananaMVGame
Is there a separate system for collision detection that somehow becomes de-synced with Mario's position? If it does, how? Does it automatically correct itself? Love the videos, and appreciate any idea on this!
There's 2 kinds of "invisible walls", and the first one is out of bounds. That means that if There's no floor under a place, That place is out of bounds. By the way, some stuff that you May think is out of bounds May not be that, for example Lava and Death barriers. The second one is a cealing that has no floor above it. Cealings extend upwads indefenetly unless There's a floor above it.
2:34 Sure... windmills...
Walls, Walls & Walls
4:24 how the hell did you slide there for a few frames?
You somehow managed to stand on the side of the pipe for six frames around 4:25.
geweldige naam man!
I like how your videos ends just end