Super Mario Bros. – Lecture 4 – CS50’s Introduction to Game Development

Super Mario Bros. – Lecture 4 – CS50’s Introduction to Game Development


[MUSIC PLAYING] COLTON OGDEN: Hey, good evening. Welcome to GD50 lecture four. This is Super Mario Bros. As seen on the slides
here, though, we’re not using the actual Super
Mario Bros. sprite sheet. This is sort of like a rip off. But I found a really
awesome sprite sheet that has all the basic tiles that
we need to get this thing working. There’s a link in the distro as
to where you can find it online. I had a lot of fun playing with it. So hopefully, maybe
if you’re curious, you can use some of the sprites in there
to go off and do your own thing. But Super Mario Bros.– the actual game
which this lecture and assignment are based off of– is the game shown here. I think everybody knows what it is. It’s probably the most
famous game of all time. But this game came out in 1985– sort of revolutionized
the gaming industry. It was the game that brought
the gaming industry from a crash in the ’70s thanks to a lot
of poor game making policies and companies and low QA standards. It basically took the gaming
crash of the late ’70s, early ’80s and brought games really back to the
forefront of people’s consciousness. This and games like Legend of
Zelda and a lot of other NES titles made Nintendo basically the
dominator of the video games industry in the ’80s and ’90s. And even today, with games
like Breath of the Wild, they’re still doing their thing. This is Super Mario Bros. It’s a 2D platformer. And what this basically means is
you control Mario, who’s a plumber. He goes around, walks sort of
looking at him from the side. He walks left to right. He can jump up and down. He’s affected by gravity. He can hit blocks. He can jump on enemies. He can go down pipes, and
there’s a bunch of levels. It was, for its time,
quite a complicated game, and it spawned numerous offshoots
and rip offs and other good quality platformers. While we talk about
Super Mario Bros. today, some of the topics we’ll actually
be talking about are tile maps– so how we can take basically
a series of numbers– tile IDs– and turn
that into a game world. As you can see here, the game is broken
up into blocks of 16 by 16 tiles. You can see the bricks and
the question mark blocks, and the pipes are even all
composed of simple tiles. And they map to IDs. And when you take a 2D table or array
and you just iterate over all of it and render the appropriate
tile at the appropriate x, y, you get the appearance of
existing in some game world, even though it’s just composed
of a bunch of tiny little blocks. 2D animation is something
we’ll talk about. So far, we haven’t
really done any animation at all in terms of at least characters. We’ll do that with Mario. He’ll have– our version
of Mario, an alien– when he’s moving, he’ll have
two frames of animation. The frames of animation–
that’s sort of like a flip book if you’ve ever used one, where
you can see individual pictures. And when you display them
rapidly back to back, you get the appearance of animation. We’ll be talking about that. Procedural level generation–
we’ll be making all of our levels generate randomly. So every time we play the
game from the beginning, everything will be completely different. We don’t have to hard code
a finite set of levels. Everything will be dynamic and
also interesting, in my opinion. We’ll talk about the basics
of platformer physics and how we can apply
that to our game world here, because we are just
using a table of tiles, each with an x, y that’s hard
coded in the game world space. All we have to really do is take
an x, y of Mario, for example, and then just divide
that by the tile size. And then we get basically
what tile that is in our array at that point in the world. And so it’s really easy to do
arbitrary collision detection based on what direction you’re going
and not have to iterate over every single tile in your world. Because it’s just a simple mathematical
operation to get the exact tile given two coordinates, since
the world is in a fixed space. We’ll have a little
snail in our game that walks around and does a
couple of random animations and will go after the player, sort
of like a little basic intro to AI. And then lastly, we’ll touch on
things like power ups and game objects and how we might be able to
influence Mario and pick those up and that sort of thing. So first though, a demo. So if anybody is willing
to come up on stage to test out my implementation
of Mario, that would be awesome. James? I’m going to go ahead into here,
so we should be all ready to go. So as soon as you’re ready,
go ahead and hit Return there, and you should be up and running. So as a part of having random levels– so currently, we have a green alien. The blocks have a random chance
in this case to spawn a gem. And so once they do,
you can pick the gem up. Either they have a gem or they don’t. You can pick it up,
and you get 100 points. As we can see, the world
is sort of shifting based on where James’s avatar
is, so it tracks the character. We have some notion of a camera. You’re getting unlucky
with the blocks so far. So you can fall down through the spaces,
so you probably want to avoid that. But if you want to
demonstrate doing it– so in that case, we collided
with the two blocks below it. The one on the right had the gem. So go ahead and just fall
down so we can demonstrate. So when we fall down, we
detect whether the player has gone below the world limit,
and then we start him back at the beginning of the game. You can press Enter, it should
regenerate a brand new world. Notice how we have random
holes in the ground. We have random tiles. We have random toppers for them. All the blocks are random. We have snails now. They’re sort of chasing after James. He can jump on top of them. There’s a lot of little moving
pieces here, but a lot of them are actually pretty simple. And I’ll show you very shortly. JAMES: Should I stop? COLTON OGDEN: Yeah, sure. That would be a great point. So thanks, James. Appreciate it. Currently, there is no
notion of a level ending. That’s part of the
piece that actually will spawn an object that the player can
interact with to just sort of retrigger a new level, basically. But the whole engine behind this basic
platformer is there, and it all works. And so our goal is seen here. Our goal in this lecture
is to demonstrate how we can get things like a character
that moves around on a screen, and a camera that tracks their
position, and tiles that are randomized. And maybe there are pillars in
the ground, holes in the ground. All of this, again– at least the tiles– are
stored as just numbers. So all we really need to do is perform
a transformation on a series of numbers. Maybe 1 is equal to a tile being
there, 0 is equal to empty space. And so just by looking at it,
we’ll see we go column by column. We can say, oh, maybe there’s a chance
to not spawn any tiles along the y column on this x of the world map. Or on this particular y, maybe
instead of spawning the ground level, we spawn a couple above it and
down so that we get a pillar and so on and so forth. And it’s just this summation
of these randomizations equals a nice little
variety of game levels. So the first thing we should talk
about really is what a tile map is. And what I’ve alluded to so far is
you can really think of a tile map as being effectively a 2D
array or a table of numbers. And it’s a little more complicated
than that depending on how complex your platformer is, because some numbers
are equal to tiles that are solid or not. So you should be able to check
whether a tile is collidable, meaning that the player or whatever
entity you want to check for can actually collide with it or Not. So obviously, we don’t want to
trigger a collision on empty tiles. We want the player to
move freely through those. But if they run up against a wall
or if gravity is affecting them, and they hit tiles below them or above
them, we want to detect a collision and then stop them based on
which direction they’re moving. And depending on how complicated
you get with your platformer, maybe you have animated tiles, for instance. So if a tile’s animated, it will
display a different frame of animation based on what timer you’re on. Really, the sky’s the limit. In this case, we’ll be fairly simple. Our tiles will mostly just be numbers
with a couple of other traits, which we’ll see later on. And this is just an example here of
a very simple map– just a colored background. We have our character, and then we can
sort of visualize all of those tiles as being just for the
sake of theory 0s or 1s. So tiles0– so I’ll actually get
into a little bit of implementation here as to how we can get
drawing some very simple tiles. So if you’re looking at
the distro, in tiles0 is going to be where we start off here. And I’m going to go ahead and run tiles0
so we can see what that looks like. So this is just tiles0. It’s a much simpler program than what
we just saw, but all we’re doing here is just a color in the
background and then tiles. Off the gate, anybody have any ideas
as to what the first step would be if we wanted to implement this? AUDIENCE: Just put the tiles in a loop,
draw them, and then have a background? COLTON OGDEN: So put the
tiles in a loop, draw them, and then have a background. Yes. So basically, if this is
main.lua in our tiles0, first thing we’re going to need
is a tiles table to store our– we’re not going to be
storing just flat numbers. We’ll be storing little mini tables
that have a number in them and ID, so we can say tile.ID
if we have a 2D table. Here, we have an empty table. We’re going to populate that. If we’re going to draw our tiles, we
are going to need a sprite of some kind. And what I did was I just chopped
out a little segment here. So this is tiles.png. It’s just literally one
tile from the main sprite sheet that comes with the distro. And then on the right side is just
transparent so that we can offset– maybe tile ID 1 is equal to solid block,
and then tile ID 2 is equal to empty. And so if we recall generate quads,
we can split up a sprite sheet into however many quads we want to. Let’s say this is 16 tiles tall– each tile– and then the whole
thing are two tiles wide. So it needs to be split
into two separate tiles. We’ll just generate quads, and
then we’ll have, recall, a table. Each of the indexes
of that table will be a quad that maps to one of these tiles. So number 1 will be
this tile here, number 2 will be the transparent
bit over here, and then that’s how effectively
our IDs are going to map into what gets drawn onto the screen. The ID is the index into our quad table. So going back into tiles0, we have
here just a map width and height. We’re just going to say
generate a map 20 by 20. RBG– we’re just going
to make it random, so we’re going to clear the
screen with a random color. And then this is the
quads=GenerateQuads. And notice that we’re
passing in tile size here. It’s good practice just to
make your tile size a constant. So our tile size in
this entire lecture– they’re all going to be 16 by 16. And so since they’re symmetrical,
we just pass in tile size TILE_SIZE. And then here is where we
actually end up spawning the map– so nested for loop. y gets 1 to map height,
x gets 1 to map width. Remember, we have to insert a blank
table into the base table that’s going to act as our current row. And then in that row, we’re
going to add a small at tiles y, because y is going to be up
here– our current row and ID. And so what we’re doing here
is if y is less than 5– meaning we’ll just set an arbitrary
point for the ground, basically. If it’s less than 5 tiles from the
top, then just make it the sky. And so sky– up here on
line 24, 25– we just set two tile IDs, as I said before. Sky is 2, so it’s going to be
on the right side of the sheet. And then ground is one. It’s going to be the very first
quad generated in the sheet. So if y is less than 5, that
ID should be equal to sky else it should be equal to ground. And so down here is where
that comes into play. We’re going to clear the
screen with our random color. We’re going to iterate over
the loop, as James said. We’re going to get
the tile at tiles y x, and then we’re just going to draw the
sheet and the quads at that tiles ID. And then recall, since tables are 1
indexed but coordinates are 0 indexed, we take the x and the
y, subtract 1 from them, and then we just multiply
them by tile size. And that has the effect of
drawing each of those tiles at their respective point
in the world and making it seem as if we have this
world– this bunch of bricks with a random background every time. Which isn’t all that interesting,
but just a little bit more variety. And so that’s the very
basic gist behind it. I mean, it’s essentially almost the same
thing as what we did in match three, where we just mapped the
individual tiles that were in the grid to indexes in the tile
sheet based on the color and variety. Only this time, they’re always
going to be in the exact same place, so we don’t have to worry
about whether their x and y are different from their grid y and grid x. We’re not maintaining
a reference to those. And so that’s static tiles. Does anybody have any
questions about how we just draw static tiles to the screen? Pretty basic stuff. The whole name behind
side scrolling game is that the tiles scroll based
on what we’re doing in the game. It can be an auto
scroller, in which case maybe you’re an airplane
that’s sort of going through a level that’s
scrolling automatically, and you’re shooting things. And you’re not really in
control of where you go. Or it can be like Mario,
where you control an avatar. And you can walk around
and jump and stuff, and the camera will
always be fixed on you. And so the scrolling is just relative
to where your character’s x and y are. So I’m going to show you
guys an example of how we can get scrolling implemented in our game. And to do that, the function that
we’re really going to be looking– at a new function– is love.graphics.translate(x, y). And so what that does is
effectively just translates Love2D’s coordinate system so
that whenever we draw something, it gets automatically shifted by x, y. And so that has the effect of the
everything being sort of skewed based on the x, y that we pass it. And so if we maintain a reference
to where the character is, we can just shift where everything
gets drawn on the screen. And that will have the effect of
it being a camera, but it’s not. All we’re doing is just
shifting the coordinate system based on some offset– x being in this case where
the players is effectively. AUDIENCE: So it changes the
whole coordinate system? COLTON OGDEN: It does. It shifts everything in the coordinate
system that you draw by the x and y. And so that will basically
affect what’s getting rendered into the active window at that time. So I’m going to go ahead
and pull up tiles1 here so we can see how this works. Let me go ahead and
first run the program. So if we’re going into
tiles1 in the distro, currently it looks almost identical. But I can move it if I
just press left or right. And so we can see here, this
is where the 2D array of tiles gets cut off here. And then it also cuts off, because
we’re only generating a 20 by 20 level. It also gets cut off at the
very right side as well. And these are details you would
normally hide from the user by just clamping the x
between 0 and the right side of the map minus VIRTUAL_WIDTH. And that will have the effect of
whenever you get to this point, it won’t let you go right anymore,
and same thing for the left side. Well, all we’re doing
right now– we’re not doing it based on the character at all. We’re just using keyboard input. So let’s go ahead into tiles1. And so the important thing that
we’re going to look at is– as I just alluded to, we’re
calling love.graphics.translate on some value called cameraScroll. It has to be a negative value,
because if we’re moving to the right up here or to the left– if we’re moving to the left, camera
scroll basically is going to decrement, so it’s going to get less. So we can say the
camera scroll when we’re going left is going to be 0 or
less if we’re starting at 0, or it’s going to decrement. If we press right, camera
roll should increase. If we want the appearance of moving
to the right or moving to the left, you actually have to translate
by the opposite direction. Because if we look at this, and if we
call love.graphics.translate positive, all of this is going to
get moved to the right. So it’s going to have the
appearance of us moving left. And if we translate it to the
left by a negative amount, it’s going to have the
appearance of us moving right. So if our scroll is positive and
we want to move to the right, we actually have to translate
by a negative amount. And so that’s why I’m calling
negative math.floor(cameraScroll). Does anybody know why we’re
calling math.floor on cameraScroll instead of just calling
negative camera scroll? Does anybody remember
what math.floor does? So math.floor will return the– it’ll basically truncate the
number down to the lowest integer. It will basically take off
the floating point value. Because we’re rendering
to a virtual resolution with push, if we basically offset the
translation by a fractional amount, you’ll get artifacting. Because it’s taking your
window and just condensing your image onto a virtual canvas, you’ll
get weird blur and stuff like that. So whenever you draw something and you
have a fractional number for something, and you’re drawing it to a virtual
canvas that’s been magnified or it’s being condensed, just
make sure to math.floor it so you don’t get any
weird blur artifacting. If you take this out and experiment
around, or even if in the distro you take it out of the player’s
position, you’ll see the player will get weird blurry
artifacting and stuff like that. So that’s why that’s there,
in case you’re curious. And so all we’re doing
here– we’re just saying if it’s equal to left, scroll
the camera left, scroll it right, or basically decrement our camera scroll
and then increment or camera scroll. And then just use the
negative version of that here. You could also just assign
camera scroll equal to positive when you move left and
negative when you move right, and then you could give it the
regular camera scroll here. But it’s sort of mentally
flipped in terms of this part. So I just made the decision to decrement
it here when we’re pressing left, because we’re going less on the x and
then more on the x when we press right. Does this make sense? Anybody have question? AUDIENCE: I have a question. Is there a corresponding
function in JavaScript and in other languages like this where
you can shift a whole coordinate? COLTON OGDEN: Is there an
equivalent function in JavaScript where you can shift the
whole coordinate system? Not in base JavaScript, probably. I’m not too familiar with CSS. There might be a CSS
function that does it. In a lot of 2D game
engines, yes, I would say. And a lot of actual 2D game engines
will have a camera object, which sort of encapsulates this behavior. Love2D doesn’t have a
camera, so this is sort of why we’re doing this– is because
it’s kind of a lower level game framework, Love2D. It doesn’t really give
you as many things right out the gate, which makes it
great for teaching these concepts. But a more robust solution
like Unity or Phaser or a lot of other game frameworks is
that they’ll just have a camera object. And you just basically give that
your x, and then you just move that. You basically tell that
to track the player– like camera.trackPlayer
or trackEntityPlayer– and that’ll have the same effect. It’s a little bit more abstract. It’s a higher level
than what we’re doing, but it’s the same exact
principle underlying. So any other questions
as to how this works? All right, cool. So that’s all we’re effectively doing. We’re just getting a camera scroll,
decrementing it and incrementing it. And then just every frame,
we’re translating everything before we draw everything. You have to do the translation before
you draw, because everything that you draw after the translation gets affected
by the new coordinate system change. So that’s scrolling. Let’s get to actually talking about
drawing a person– an avatar– more than just a set of tiles, since
that’s what the game revolves around. If we look at character0, this
would be our first example here. This is just going to be
a very simple example– charactr0. You guys probably know
how this works already. All we’re doing is just drawing
a sprite to the screen– so just love.graphics.draw. We’re getting quads from a tile sheet. I believe it’s in the slides. The actual sheet is here. So we have this little guy– several frames of animation. It’s 16 wide, 20 tall,
and we just take a quad. We split it up into quads first. So we know that it’s 16 wide by
20 tall, so we just generate quads on this image by 16 and 20. And then in this
example, all we’re doing is taking the first frame, which
is quads1, and just drawing that. As you can see here, we have
a bunch of different things. We have like a crouching
state, and we’ll get to more about
animations in a little bit. But here, we have him
climbing up a ladder. But you can see all
these different frames. We’ll end up showing how you
can play them back to back and get different animations. But for the sake of this
basic example, all we’re doing is just rendering the very first frame. And we can see that– let me make sure I’m in the
right file, which I am– we are getting the character
sheet here on line 43 and 44. And then we have to give him an
x, So characterX, characterY. In this case, we’re just
setting him above tile 7, so we do 7 minus 1 times TILE_SIZE
because tiles are 1 indexed but coordinates are 0 indexed. And then we just subtract
the height so that he’s right above the tile instead
of right at the tile. And then down here, we
do a love.graphics.draw on, as I said before,
just characterQuads 1. Just a very basic hard coded example. Any questions at all
as to how this works? So now let’s say we want him to move. What do we need? What’s the next step if we
just wanted him to move? AUDIENCE: Give him an x and y? COLTON OGDEN: Yes, give him an x and y. So yes, so he does have
an x and y already. So if you look at–
am I in the right one? So if you go to character0,
this is character0 still. We have given him an x and y already,
but there needs to be another step. What’s the other step involved? So if we wanted to move, we need
to check for keyboard input. And then we need to take his x– we’re just going to move
him on the x-axis for now. We basically need to take his
characterX variable up here, and we need to modify that. We can basically do the same thing
that we did down here in love.update. Previously, it was on
the coordinate system– love.graphics.translate. We modified the camera scroll. We set that equal to scroll
speed times delta time. We subtracted or added it. In this case, what we’re doing
is we have a new constant called CHARACTER_MOVE_SPEED, and we’re
just doing that exact same operation but on characterX
instead of cameraScroll. So the end result of that is
that we have the character here. And then we can move him left
or right, and he go off screen. Now, there’s a couple of things wrong. What’s wrong? What are some of the things that
are wrong with the scene right now? AUDIENCE: The camera
should move with him. COLTON OGDEN: Camera
should move with them. AUDIENCE: No animation. COLTON OGDEN: Does not have animation. Those are probably the two
real things that are wrong. So camera does not track him,
which is an important thing. Obviously, we want to be able to
maintain a reference to our character, unless we’re at the
left edge of the screen. If we’re at the left edge of
the screen, this is actually OK. And that’s part of the
distro– is we clamp the x so that it doesn’t go past the left edge. But if we’re beyond the middle and
not to the right edge of the screen, it should be moving along
with him and vice versa. And then he needs to
animate, so his sprite needs to change every certain number
of seconds whether he’s moving. And it has to be only
when he’s moving, right? If he’s standing still, you
can have an idle animation. Some characters will tap their
foot and do stuff like that. But let’s say for the sake of
this example we want him just to stand still when he’s idle. And we want him to have an actual
animation when he’s moving. We need to take care
of these two pieces– three pieces if you count
the idle animation part. So let’s go into character2 and
take care of the first part, which is tracking him. So let me go into character2. Let’s run it first so we
can see what it looks like. So now the camera is basically
affixed to the player. In this example, we don’t take
care of the left edge issue. In the distro, that’s fixed. But we have the basic side scrolling
mechanic– take a character, follow him. How do you think that
we’re accomplishing this? Yes. AUDIENCE: Translate the
drawing against the characterX? COLTON OGDEN: Yes, exactly. And it can’t be exactly
the characterX though, because if it is, then the character
is going to be on the left edge, right? So we need to offset our
x that we translate by. We need to basically translate by
his x minus half the screen space plus half the character width. And that will have the effect of
translating it but always keeping that offset half a screen width away
from the player, if that makes sense. And so what we’re doing is in
character– this is character2, right– character2, we’re still
doing the same thing we did. Actually, that’s the wrong file. We are modifying characterX here. So same thing we did before– multiply the move speed by delta
time and either add or subtract it if we’re pressing left or right. But also, here–
reintroducing camera scroll. And we’re setting it to, like I said,
characterX minus VIRTUAL_WIDTH divided by 2, half the screen,
and then positive offset of his width divided by 2 so that
he’s perfectly right in the center. Because remember,
characters’ coordinates are set by their left, not their center. And then we just do what we did before. We translate the scene
based on cameraScroll, and we render him at characterX
characterY using math.floor to prevent him from being at a
fractional point in our world space and then it being blurry and artifacted. And that’s sort of it in terms of how
we can get tracking over character. And if you wanted to
track along the y-axis, you could do the exact same thing. Maintain a cameraScroll
x and a cameraScroll y– so keep them separated. And then you would just translate here. So we’re passing in 0,
because we don’t want to track along the y-axis necessarily. But all you would need to do
is pass in your y cameraScroll. And then you could do
it based on characterY and whether or not
they’re above the ground or past a certain point in the sky. So any questions at all as to how
the camera tracking is working here? All right. So we took care of one issue,
which was the lack of tracking. But there was one other issue,
which was he’s not animated. All he’s doing is just moving
sort of like M.C. Hammer– or is it M.C. Usher? M.C. Hammer? I forget. He’s doing that. He’s not doing anything. We need to actually animate him so that
he looks like he has some life to him and that you can also
differentiate importantly between two separate states. He can be idle, he’s not
moving, and he can be moving. So we should have some
sort of visual feedback as to what’s currently going on. So anybody know how we
can go about implementing an animation for our character? What are the pieces that we’ll need? AUDIENCE: I guess if
he’s moving right, then call a function and a render
that looks through some images? COLTON OGDEN: Yes. So if he’s moving right,
then have a function that sort of loops through some images. That is effectively
what we will be doing. We have a class called Animation,
which I’ve introduced here. And all it basically does
is keep track of– you pass it in a table, which
has the frames of the sheet that you want to animate over. So we can just pass in– let’s go ahead and take a look here. And I referenced the slide
earlier, but all of these are 1, 2, 3, 4, 5, 6, 7, 8, 9,
10– however many there are, you just pass into the animation. Let’s say he’s on a ladder. So let’s say this is 1,
2, 3, 4, 5, 6, and 7. You say, the frames are
going to be 6 and 7, so those will just loop
left to right, starting back at the beginning when it’s finished
And then you give it an interval. So say I want the animation to
happen this fast in terms of seconds, so I want it that maybe
happened every 0.2 seconds. And so that will have the
effect of every 0.2 seconds, it’ll keep track of a timer. So have we gone over 0.2 seconds? Start at 0 and then add
delta time to it every time. If we have, increment what our
current frame of animation is. So our current frame is this one,
and then 0.2 seconds elapses, it’s going to be this one. And then 0.2 second elapses, and we
need to loop back to the beginning. So we’ll end up using
modulus to take care of that as we can see in the Animation class. Basically, that’s all done here. So if we have more than one frame of
animation, recall it gets a def here. So we get frames, we get an interval,
get a timer that’s initialized to 0, and then get a current frame. We’ll say the current frame is 1. And then as long as we
have more than one frame, there’s no point in looping
over or trying to animate any animation that only has one frame. And we can, of course, have
animations that only have one frame. Idle is only one frame of
animation, as we saw here. That’s only one frame. We don’t need to do any
sort of logic to say, oh, what’s the next frame, because
there’s only one frame. But if we were to look at character3,
we can see two frames there. And then that’s just
one, frame he’s idle. And when we move left, he
moves in that direction. Anybody recall how we can
get him– because obviously, we saw the spreadsheet
just a second ago, and there was only one direction
that the sprites were facing– how we can get him to look
that way, even though there’s no sprites for him to look that way? AUDIENCE: Flip it on the axis. COLTON OGDEN: Flip it,
so love.graphics.draw. Recall you can pass in a negative
scale factor on whatever axis you want, and that’ll have the result of
flipping it along that axis. That’s all we’re doing. So this is the default frame,
so we’re just drawing it. And then we have to keep a reference
to whatever direction he’s facing. And if his direction is
equal to right, we’ll just draw that frame and then
loop and process the animation. If he’s facing left, draw it, but also
perform a negative 1 transformation on the x-axis. And just like that,
we have that working. So all we’re doing– just keep a timer. And then when the timer goes over our
interval, just increment the frame. And then use modulus
to loop back over it– back to starting at 1. And that’s all done
here on this line 28. And so you can look
in there a little more if you want to get a handle
on how the math works, but it is just a simple sequence of
iterating over a collection of frames based on a timer. And that has the effect– just like
a flip book, as I said earlier– of our character having an
animation and having some life. So any questions as to how
this animation class works? AUDIENCE: The render is
in the Animate class? COLTON OGDEN: So no. The render is not in the animate class. So the render is– I realize I didn’t show
any actual main here. We have two animations here,
which was just the idle one, so we’re just passing in one frame. We’re going to give it an interval of 1. It’s not going to really matter, but
just for the sake of consistency, we’re giving it an
interval of 1– arbitrary. And maybe we want to
change his animation later. So by having an interval here,
we won’t forget to add one later. Moving animation– recall 10 and 11. So it’s toward the end of the sheet– the two walking frames. Interval here is 0.2 seconds. We need a current
animation to render him, and then we keep a reference to
whatever direction he’s looking at. So if he’s looking to
the right, we’re going to reference this in the
love.graphics.draw at the bottom. And that’s what we’re going to
use to perform the sprite flipping along the x-axis. Maintain a reference to that. And then down here, the part that
we actually reference the animation is on line 150 if you’re
looking at character1. Or is it character2? Sorry, character3. If you’re looking at
line 150 in character3, we’re using
currentAnimation:getCurrentFrame(). So the class will actually just tell you
whatever the current frame of animation is, because it keeps a reference to
what frame it is based on the timer and how much has elapsed. AUDIENCE: So the class is generating
a different frame real time and plugging it in there. COLTON OGDEN: Yeah. It’s maintaining a reference to
whatever the current frame is, and it’s in the table of frames
that it got when you gave it the definition up at the top here– lines 51 to 58 where we
create the two animations. Basically, it maintains a reference
to which index in this frame table we’re at. So if 0.2 seconds has elapsed, we
start at 1, and then we go to 2. And then we’ll go back to 1. And so it’ll just basically
return frames, index. And frames index 1 is
10, frames index 2 is 11. And so the function is getCurrentFrame. So characterQuads,
currentAnimation, getCurrentFrame. And then here, because we’re
performing an origin transformation– so that’s another thing to consider
when you’re flipping sprites. When you flip a sprite,
it actually flips along whatever its default origin is. And the default origin of any
sprite is its top left corner here. So if you flip something
along its x-axis, it’ll appear here instead
of just flipping in place. So you actually have to set
the origin to its center when you do any sort of in
place flipping of a sprite. So you’ll notice in the
code when you’re looking at it that we have plus CHARACTER_WIDTH
divided by 2 and plus CHARACTER_HEIGHT divided by 2 on these two here. So we shift where it gets drawn,
and then we shift its origin offsets which are here on line 160. So if you look at
love.graphics.draw, you’ll see it has a lot of optional arguments. And these two at the bottom are
the origin offset arguments. And so these only really
come into play when you do some kind of flipping
of a sprite on an axis and you want graphical consistency not
to have it flip one way or the other. Sometimes that’s the effect you’re
looking for, but in this case, it’s not. We want him to literally
stay in the exact same place. So to flip a sprite in
the exact same place, you need a set its origin to
its center, not its top left. Does that makes sense? OK. And also here, 0 is the rotation here. So it’s sort of required
if you’re going to add this many arguments to the function. But we’re testing if direction is equal
to left, we want to flip by negative 1 on the x, else just give it 1. So 1 just means default
transformation, so no flipping. And then we don’t flip on the y
at all, so that will always be 1. And so that’s in a nutshell how you can
get your character to animate and also stay in place when you animate it. So any questions as to how animations or
the origin offsets or any of that work? OK. So we did talk about animations. The last thing we’ll talk about
for the character is jumping. So if you recall from Flappy Bird,
how can we get our character to jump? What are some of the pieces we need? AUDIENCE: Key press,
and then the y goes up. And then we have to have gravity. COLTON OGDEN: Yep. So key press is one thing
we need, so check for space is going to be the default key. y goes up, and then check for gravity. So not only do we need y,
but we also need delta y. We need velocity, because gravity
is a transformation on velocity, not strictly on position. So if we go back to
character4, this is sort of a hackish way of
implementing gravity, because we haven’t actually
incorporated tile collisions. And I’ll defer most of the
implementation for that as to the distro, and I’ll
go over with you guys. But right now, we have the
exact same thing we had before, where we have tile scrolling. But if I press space bar, I
go up, and then he comes down. And notice that he has
an animation as well. He has a different frame. So if he’s jumping, he’s
got a little jump frame. So that means now we
have three animations. We have an idle animation,
we have a moving animation, and then we have a jump animation. So effectively, we have
three states as well– idle state, moving
state, and jumping state. Four states, actually. And also, I noticed a slight bug here
where if you’re still in the air, his frame doesn’t change. So it actually probably
should stay to that frame, even if he’s standing still. But I guess it doesn’t matter too much. We also interpret it as a feature. But he’s got a couple of
states when he’s in the air. There should be two states here. One is jumping state,
and one is falling state. And do we know why the two being
different is an important thing? So if we think about Super Mario Bros. and we think about the differences
between jumping and falling, what are some of the things that change
based on whether Mario is jumping or whether he’s falling? How does he interact differently
with the environment, I should say? So if unfamiliar, Mario– when he
jumps, he can actually hit blocks. So if he’s below a block
and he hits a block that has some sort of behavior in it, it
will trigger whatever is in that block, whether it’s a coin or whether
it’s to destroy the block. And if he’s falling, recall if he
lands on top of an enemy like a goomba, he’ll destroy the enemy. And so we need to distinguish
between these two states. Because when he’s jumping,
he’s not able to– when he’s actually going up,
he can’t attack the enemy. And likewise, when he’s falling
down, he can’t destroy the block. So even though he’s
jumping up in the air and the gravity is
applying a transformation and it all looks like one state,
there’s actually two important changes in his state that are relevant. And that’s something that
we’ll need to pay attention to, and it’s in the distro. He has a falling state
and a jumping state. Even though they share
the same animation, they have different behavior. So let’s go ahead and look at
the character4 distro here. So what I’ve done here is I’ve
added a delta y for the character. So just like in Flappy
Bird when we press space and we made our delta y
go up to negative 50– so instantly shot up pretty
high, because that was getting applied every frame. Same thing here. Once we press space, we’re going
to change delta y to negative 50 if we go down to right here. So if the key is equal to space, I have
it in the love.keyPressed function. Since we’re doing all this in
main.lua just for illustration, things are a little simple. If key is equal to
space and his delta y is equal to 0, what would
happen if we didn’t check to see if delta y was equal to 0? AUDIENCE: We double jump in the air. COLTON OGDEN: Yep. We’d be able to jump infinitely,
so we have to do a check for that. We set his dy to JUMP_VELOCITY. JUMP_VELOCITY is a constant up top
on line 29, which is negative 200. And then gravity is equal to 7. And so what we do is we set it
to negative 200– his delta y– as soon as he jumps. And then every frame down
in update, we basically increment his delta y by gravity. And then we increment his y
by delta y times delta time. And so it’ll have the effect
of when he’s in the air and he’s got a negative
velocity, it’ll actually start becoming positive and
positive until it is positive, and then he falls back to the ground. And then the hack that I
was referring to earlier– since we don’t have collision detection
implemented in this example yet– is we’re just basically
checking to see whether he has gone below what we set the map’s floor. And if he has, then set his position,
first of all, to be above that tile here on line 133. And then set his delta y equal to 0. And that will allow us
then to hit space again, because his delta y will be equal to 0. AUDIENCE: So I didn’t
see [INAUDIBLE] on it. Looks like there’s always gravity, then. COLTON OGDEN: There is
always gravity, something I realized shortly before lecture. But all you would really have to
do is, I think, if character dy– Yeah. You could easily take that out of
there– just an if statement around it. AUDIENCE: It’s just a
waste of resources, right? COLTON OGDEN: It is. I mean, it’s not expensive,
because all you’re doing is incrementing a variable
by a certain amount. If anything, if you’re introducing
an if condition every frame, which is probably the
same if not actually more. I think a branch is more
CPU than just an assignment. I’m not entirely sure about that. AUDIENCE: Interesting. COLTON OGDEN: Yeah. In this case, it doesn’t
really have any side effects. But it’s a good thing to notice. But now notice that we can just
sort of walk along the floor here, because there’s no collision detection. We’ll talk about how we implement
a collision detection soon. So one thing that we’ll start talking
on– and we’ll take a break fairly soon– is procedural level generation. So I am a big fan of
procedural level generation, and platformer levels are
actually fairly easy– at least in a simple sense– to procedurally generate. And so like with match
three, all we basically did was just loop
through our grid and just say, oh, get a random
color and a random variety. And then with the assignment, it
was a little bit more complicated, where you actually had to check to
see whether you were on level one. And then if you weren’t,
then your variety should be maybe a certain amount
depending on how far along you’ve progressed in the game. With a platformer
level, we have to think about how we can take that grid of tile
IDs and think about it mathematically. How can we get the results of a level,
but make it different every time– introduce some variation, right? And so the solution that I
found that makes the most sense is going column by column. So here, we just have a
bunch of– this is just a very simple perfect screenshot
to illustrate a very simple way of generating the level. But recall, if we just think about
these tiles here– these empty spaces– being a 0 and these being a 1, it’s
sort of almost like binary in this case. We could just fill the entire thing
with 0 first, just assume empty space. And then we could just column by
column go down and just have a chance every column. OK, do I want to generate a ground here? If I do, start at the
ground level and then just generate earth tiles all the way down. And then go to the next x position,
do the same thing, do the same thing. And then maybe every column of
the world that you’re generating, you also have a chance to
generate a pillar like this. So if generate pillar is
true, then I want to spawn– instead of starting the ground
here, I want to start it here. And then maybe you
have a flag that says, OK, not only do I want
to generate pillars, I also want to generate chasms– just empty space,
obstacles for the player. Because if he falls down– it
goes below the world space– it should be game over. So in that case, you just
say if generate chasm– make math.random 10 or whatever it is– then just go to the next x. Don’t even do anything. And that will have the
result of generating a chasm. And so little piece by piece– doing small things like
that has the net effect of generating a lot of
visually interesting, dynamic, and random levels. You never know what to expect. And this is a very basic example. You could go infinitely far with it. However many ideas you
have in terms of how to create obstacles and interesting
levels and scenery for the player– you could absolutely implement that. AUDIENCE: How do you handle if
there’s a platform to jump on? You have to have that consistency. COLTON OGDEN: Yeah. So if it’s a platform, it depends on
how you want to implement platforms. And actually, I did a
seminar on Super Mario Bros., and we did platforms as tiles. In this case, we’ll have
blocks that are actually what we’ve denoted as
game objects– which are a little bit different than tiles. Because they can have arbitrary
sizes, and they don’t necessarily have to be affixed to the world grid. But if you were to treat a
platform that was, let’s say, two tiles wide as
tiles, all you would do is just basically have a flag up
here that’s like, generate platform equals true or whatever– AUDIENCE: And then turn it off after– COLTON OGDEN: Turn it off
after however many iterations. You also need the size of it. You’ll need a flag that’s like
platform width equals however many, and so you’ll just keep a counter. It’s like current platform
tile equals 1, 2, 3. And if it’s equal to width, then
you don’t generate it any more. And that has the effect of
potentially colliding with pillars if you don’t account for that. So you can also in your logic say, if
I’m generating a platform right now, don’t generate a pillar. But you could generate a
chasm, because the chasm doesn’t interfere with your platform. If you don’t have platforms as
tiles– if they’re different objects– then you don’t have to do it during
the actual world generation phase. You can just test. You can just create a game
object that’s a platform. Depending on how complicated
your algorithm is, maybe make sure that it’s not next
to a pillar when you generate it. And you could just do that by getting
the tile here and then looking at the next four tiles– something like that. We don’t do platforms in this
example, but it’s something that you could pretty
easily do with tiles. And slightly more difficult
but also still fairly easy to do with game objects, which
is included in the distro and which we’ll touch
on in a little bit. Let’s see, we’re at level– oh, another couple of things
that I wanted to show before we actually start getting into the
code for how to generate levels. This is the sprite sheet for this whole
project, which is a really cool sprite sheet that I found online. It’s in the spirit of
platformers like Mario, and it’s got a nice little
mockup here on the right. So I encourage you to
take a look at that and just maybe get some inspiration
and see all the different cool stuff. Tinker around with it if you want to. But as you can see here, there’s a
couple of pretty prominent things. We have a ton of tiles. These are all tiles here– different tiles and variations. And then we have a ton
of these toppers here. And so what really helps
this whole demonstration of generating these levels
is the fact that we have so much visual content to work with. And so here, again, are the tiles. Here are the toppers. And then when you take the
two together and then you also have these random backgrounds– these are toppers here,
the top of the tiles here. It’s incredibly easy to just have
a sheer abundance of visual variety and interesting things in your
game levels without even– and the algorithms here are very simple. All we’re doing is just checking
to generate pillars and columns. I know. I thought it was really cool and helps
illustrate the importance and power of this whole procedural approach
to creating the levels for this. And there’s actually not that many
games that take advantage, I think, of procedural level generation
in the platformer genre. Plenty of games like
Minecraft and Terraria– Terraria is a great platformer
that is an example of that. But I don’t think I’ve seen a
really good Super Mario Bros. game that does something like that. Let’s see. What time is it? 6:23. Let’s take a five minute. And then as soon as
we get back from that, we’ll start going into how we actually
can implement the procedural level generation in more detail. All right, welcome back. This is lecture four. And before we took a
break, we were talking about procedural level generation
in the context of platformer levels. So recall, here are just a few examples
that I took pretty quickly of my code. And you can see they have different
backgrounds, different tiles. Sometimes we have chasms,
sometimes we have pillars. We’ll be talking about a few ways
to do the tile version of that, because there’s two levels here. In the distro, we’ll see there are
also things like bushes, for example. We can see in the top
middle there the purple– well, I guess those little purple cacti. And the one right below that, there
is a pillar with a yellow fern on it. Those are separate
objects from the tiles– game objects. But the actual tiles themselves we’ll
dig in here a little bit as to how to get those generated. So the first thing we want to look at,
level0, is just some flat levels– so just basically what we’ve already done. So I’m going to go ahead
and go into level0. And then if we see here, we
have a simple flat level, just like we did before. Now the tiles are different. And if I press R, they’re
randomly generating every time. So you can get a sense of just how
visually diverse this generation looks. Oh, I think that might
have been a bug earlier. I’m not sure. Haven’t seen that yet. But we can see here, I’m pressing R. All I’m doing is taking the
array of tiles that we have, and I’m assigning it a
tile set and a topper set in the case of the scope
of this generation. So recall that the topper is
just the top layer sprite, and the tile set is
the tiles underneath. Anybody want to just suggest how I’m
rendering the topper versus the tiles and what’s going on there? AUDIENCE: You’re just pulling it
from part of the sheet, right? COLTON OGDEN: Yes. Yeah, in a nutshell, I’m
just pulling the toppers from a different part of the sheet. Any idea how I’m storing
information– what’s being stored here to get it to render like this? AUDIENCE: Maybe you just need to
store the position of the topper and know that everything
else is below that. COLTON OGDEN: Yes. So you could store the
position of the topper and know that everything
else is below that. That would work for a flat level. I don’t think that would be reliable
for a level that has pillars on it, because the pillars are a higher
elevation than the ground. And then there’s also
chasms and stuff like that. So what’s going on
here actually is we’re storing a flag in the tile that says
whether or not it has a topper on it. And if it has a topper, then
we render not only the tile, but as soon as we render the
tile, we also render the topper. And I won’t go too deep
into the code here. But what we’re doing to get all these
different tile sets and topper sets too is we have to take all of these
tile sets– these collections of tiles– and divide them up, right? We have to know that if we want
to render the entire level in tile set one, then we should basically
take this into its own sheet– its own table– this into its own
table, this into its own table, going left to right actually. And we have basically
four way nested loop. So we go every set on the
x by every set on the y. And then within each of those, we want
to look for every tile along the x and every tile along the
y therein and split up the tile sets so that we can
index into the individual quads. So in the actual code, I won’t
go too deeply into it here. But I’ll show you where it is if you’re
curious to look into how we do that. It’s in Mario in Source, util.lua,
which is recall where we before stored our generateQuads function,
which does a simple split on a tile sheet along its x and y based on
whatever width and height you pass in. We have in here also a generateTileSets
function, which takes in the quads from a generateQuads table. So we first generate quads on
all of this or all of this. So we have every single frame of
this divided by 16, which is– I don’t know how many that is. 6 by 5 times 10 by 5, 10 by 4– that many quads, so thousands of
quads, I think, if not hundreds. This I’m pretty sure
is thousands of quads. And then we take that
and then divide it using the number of sets along the
x-axis, sets on the y-axis, and then the size of each tile set
along the x and size along the y. We basically divide it using
a four way nested loop here. We basically just divide it up. And then instead of
doing a generateQuads along the entirety of
the picture, we just basically do a 2D slice of
that quad table we get back from the first generateQuads call. So I encourage you to look in
here and experiment with that. You don’t need to necessarily know
how it works for the assignment. But that’s how we can basically
take a giant sheet like this and easily integrate it into our code. We can just swap in and
out whatever active tile sheet we want to work with,
assuming that everything is cleanly laid out like this, which
is on the part of you or your artist. You want to make sure that everything is
conducive to programmatic organization. Had things been scattered
around in a very awkward way– maybe things were zig zagged or there
were weird spaces or something like that– we wouldn’t be able to do something
as clean as what we did here in util.lua with just 63
minus 20 lines of code by getting each individual tile set. So that’s an important
consideration if you’re looking at creating
assets for your project and you want to do some programmatic
hot swapping of your tile sets. Let’s make sure we’re in the right– we’re not in the right
example here, so we’re going to go into level0 into main. And we have constants now for all of
our tile sets and what the height is and how many they are wide by tall. We do it here. We get our regular quads
from our tiles and toppers, so these are just literally every
single tile within that big tile sheet put in one table. And then we just divided
up into tile sets and topper sets here with
generateTileSets function. And then we get a random tile set
and a random topper set here– math.random, number of tile
sets, number of topper sets. And then at the very bottom also,
we have a generateLevel function– 223– which is going to be built
upon in the next two examples. Level0 is just a flat
level, so it’s actually exactly what we saw before, which
was just if y is less than 7, ID should be equal to sky or ground. And then this part is
actually what I was alluding to before with
the topper, because recall we need to store a flag in a
tile to render a topper or not. And it should be whatever the top
tile is in the level on the ground. In this very simple flat
thing, we can always assume it’ll be the same y level. In this case, if it’s
equal to 7, then topper should be true, otherwise, false. So every tile along y 7 is going
to have topper equals true. And this comes into play up here. If we do love.graphics.draw(tilesheet),
we have not only just tile.ID as we did before, but we have tile sets
indexed into tileset now. So remember, tileset got
a random value between 1 and however many tile sets we had that
we spliced out of our massive tile sheet. Now, we just index into that,
and then we index into tile.ID. And tile.ID will then be whatever our
ID is but relative to that sheet, not the whole entire sprite at once. And the same thing for topper. We have a topper set,
we index into topper sets here at the topper set
that we got, and then that’s where we’ll have the collection of
tiles that form that particular set. And so the two are completely separate. They can be one random color tile with
one random topper, but it’s consistent. It’s global. We have one topper set and one tile
set that are active at any one time. And if we press R, which
I did up here, then we just reset them to random
on line 139 and 140. Tile set gets a new random number,
topper set gets a new random number. It has the effect of– we can just walk around and
then generate random sets. So pretty simple. And recall again, topper is– because the tile that we’re standing
on is y seven, topper equals true. So in that case, that particular top
layer is always going to have a topper. And it gives us a nice
little bit of visual variety, because it actually makes quite
a bit of a difference having a topper versus no topper. And you can also just not
have a topper and consider that a permutation of the toppers
times tiles, like procedural algorithm. That’s flat levels. Does anybody have any
questions as to how this works or anything that we’re doing here? OK. So things are a little
flat, a little boring. The next step will be actually
introducing one of the things that we can see here in our little
collection of sample levels, like this pillar right
here in the very middle. Does anybody recall how we go
about spawning a pillar as opposed to just flat land? AUDIENCE: For that column, just put
some more dirt down or more tiles down. COLTON OGDEN: Yep. So for that column, just put more tiles
down instead of just the ground level. That’s exactly what we’re going to do. So I’m going to go ahead
and open up level1 and main, and I’ll run the example here as well
just so you can see it looks like. So here we have quite a few. And notice we haven’t
implement collision, so we’re still walking through them. But they’re just random. Their random amount is
up to taste, really. Right here it’s pretty
common, so it might be worth lowering the amount a little bit. If you wanted to, you could also maybe
have a flag that says, spawn pillar, and maybe you want a pillar width. You could have anywhere
between one and three tiles. And if its width is greater than
1, then just loop over a few times and just draw that same height a few
times as opposed to just one time, and then set the flag back to false. A lot of things you can do with it. And also, they’re a little tall here. For the main distro, I ended up
making them a little shorter. But we’ll see how we do
this in the code here. It’s going to mostly be down
in our generateLevel function. So what we’re doing here–
go ahead and hide that– is we have basically this code here– line 227 to 236. So all we’re doing here is just
filling our entire thing with just sky. We’re just setting the
entire thing to empty. And now we have a fully
populated 2D array. All we need to do in order
to change a tile– we don’t have to worry about insertions
or adding too many tiles to our array. All we can do now is just directly
change whatever tile exists there. So all we need to do is
starting on line 239, we’re going to start doing the column by
column iteration over our entire level and deciding whether we should
generate pillars or not. And we’re always going
to generate ground. So here’s the flag spawnPillar. And if it’s equal to 1, this is going
to basically be assigned to spawnPillar. So math.random(5)==1. We have a 1 in 5 chance
of spawning a pillar. If we just want a pillar, then
pillar gets equal to 4 from 4 to 6– so y gets 4 to 6 effectively– tiles at pillar x, ID ground. And then here’s where we
set the topper, recall, because now pillars can be the
top most tile on the surface. But they’re above the ground level. So we just basically say, when
we’re generating a pillar, if pillar is equal to 4– which is
the very first tile that we start at– then set topper equal to true here. Otherwise, set it to false. So that’s how we can get pillars to
also have toppers and then in this case, we’re not generating any chasms yet. So all we’re going to do– once we’ve generated a pillar
on that particular column, we’ll just say ground gets
7 until the map height– so towards the very
bottom of the screen. And then we’ll just set it to ground. And then topper– in this
case, we’re going to make sure that we’re not spawning a pillar. Because if we don’t
check this, then it’ll also spawn a topper where
the pillar meets the ground, and it’ll look a little bit silly. And then we also want to check
that ground is equal to 7. And so all together, that has
the effect of this behavior. And so if we didn’t check
for that spawnPillar, we’d have a topper right
below our feet here too, which looks graphically strange. And also, you can see– emergently, we’re getting
double width pillars. And that’s just kind
of a natural byproduct of a lot of these randomizations. A lot of these procedural algorithms– they’ll generate outcomes
that you might not necessarily have anticipated, which
is kind of a cool thing. You didn’t necessarily program it to
have pillars that were two tiles wide, but just the nature of randomization–
that’s just what you get. And that’s another exciting thing
about procedural level generation is that it can surprise even the
person that wrote the algorithm. It’s really cool, and it saves
you work having to create levels. So that was pillared levels. Chasm levels– who can tell
me how we can do chasm levels? AUDIENCE: You just skip a column. COLTON OGDEN: Yep. you skip a column. So at the very beginning,
all we can just basically say is, do I want to generate a chasm here? If I do, just skip. Go to the next iteration of the loop. And so we’ll take a look at that. As simple as it is, because Lua
doesn’t have the notion of continue– this will be a refresher,
because I believe this was in one of the assignments– it has a goto statement. So basically, same code as
before, starting column by column. x equals one until map width. We have a 1 in 7 chance– just arbitrary. And this should ideally– if you’re engineering an entire
large game or application, this would be called
SPAWN_CHASM_CHANCE probably, and just set that to seven somewhere. But we’re just setting it to 7 here– just a static magic number, but
magic numbers are generally bad. Goto continue– and so continue is
here at the very bottom of the loop here, which is this for x=1, mapWidth. So it will have the effect of skipping
straight to x equals 2 if this at 1, for example. A lot of languages just
simply have continue. Lua does not have continue, so this
is a community established tradition for implementing
continue-like behavior in Lua. You create a label via double colon
with a name and then a double colon, and then you just goto it. And so that’s as simple as
it is for generating chasms. And so if we go to level2
and run that, we get chasms. And so now we’ve got a little bit
of interesting visual variety. It’s not spawning a ton
of chasms in this example. It spawned one so far. There’s another one. And then sometimes just
emergently, you can get two. See, there we go. We get some interesting
obstacles as a result. It almost looks as if someone
intentionally did that– almost. I would probably, like I said,
shrink the pillar size a little bit. It’s a little tall. That’s that. That’s basic procedural– in
the context of platformers, that’s the mental model
for how we can start thinking about generating obstacles. And there’s a lot of different
directions you could go. Let’s say maybe you wanted
to generate pyramids. I mean, it’s a common thing in Mario. There will be steps,
[INAUDIBLE] set for it. The same implementation
would basically happen here. It would be a little bit
different, because you’re doing it on a column by column basis. But you’d effectively
just maintain a reference to something like step
height, and then you would say generate stairs is true here. And then you would just
set step height to 1. So then you add a tile here. You would go from ground level up
until step height, generate a tile, go the next one, and then
increment step height to 2. And then do from ground
until step height– tiles go up. So 1 and then 2 and 3 until you’ve
gone to stairs width, in which case you stop generating stairs. That’s this principle
behind how you could do something a little more complicated. Or pyramids– same exact
thing, pyramid width. And then you just go until
pyramid width equals– or we’re at pyramid width
divided by 2, make it go up. And if we’re higher than
that, make it go down. And then you have the effect
of the pyramid approach. Yeah. AUDIENCE: Where are
you putting the column generation if it’s a [INAUDIBLE]. It’s not in the play state. COLTON OGDEN: In this
case, it’s all in main.lua. But in the distro, it’s going
to be in levelmaker.lua. So we’ve broken out all
of this functionality into just how we did Breakout. We had the same sort of
thing– level maker– and it just has levelMaker.generate. And then you give it
a width and a height, and it will generate an
entire level for you. AUDIENCE: An entire level,
but it has to continuously– oh, you generate it all at once? It doesn’t generate as you walk? COLTON OGDEN: The question is, does it
generate continuously or all at once? It just generates all at once. So you could implement a– if you wanted to do an infinite runner,
the way you would do that is you would break up your level into chunks. And with infinite runners, usually
you can only move in one direction. So as you go right, your levels that
you’ve generated before– they get discarded, so you avoid
memory overconsumption. What you would do is you
would just generate a chunk– maybe a 100 by 20 level. And then you would go
through that, through that. And then when you get to level end
minus maybe like five tiles or 10 tiles, you would generate another one,
append it, put it to the right, and then you would just go
from the left to the right. And you probably would need
some sort of semi-fancy code to splice them together
once you’ve generated them. Alternatively, you could
just always pad your– no, you probably wouldn’t want to pad. I would probably just splice them end to
end and then get rid of x equals 1 100 or however many on the left once it’s
gone past the left edge of the screen. In this case, to
summarize, it’s all static. But you could very easily– not easily, but you could very
well make it an infinite runner. Yeah. AUDIENCE: So we’re rendering the entire
level, but we just can’t see it all? COLTON OGDEN: The question is,
are we rendering the entire level, but we just can’t see it all? The answer is yes. Currently, in this implementation,
we’re just rendering the entire level– so tile by tile is getting
drawn to the screen. For small examples like
this, it’s not a concern. But for a large level– like if we
did a Terraria level, for example. Terraria’s thousands and thousands
of tiles wide by probably 1,000 or more tiles tall– you want to render only a chunk,
only what you can visibly see. And for that, you could use your
camera offset and then just render from one tile to the left and above
that to one tile below the bottom edge of the camera and to the right of it. Just render that subset of tiles. So you just need a for loop to
iterate over a small section. AUDIENCE: So you can
kind of make an array of what the map’s going
to look like and then just render only slices of
the array that you can see. Is that right? If you put a multi-dimensional array
and then you just go through it and render as you go–
is that the thought? COLTON OGDEN: Question
was, you just have a multi-dimensional array
of tiles for your level, and then you just render it as you go. The answer is yes. You would have your overall tiles– your big 2D array of 100 by 20 or
however many thousands of tiles. And then based on wherever
your camera is rendering, it’s just a for loop within
that just of a nested amount. So maybe your player is
at x 30 plus 6 tiles. So you would just render from
30 tiles to maybe 45 tiles on x and maybe 10 to 20 on the y– just that chunk. And it’s just relative
to where your camera is. You’re always rendering
just a small little– basically, it is effectively
a camera at that point. It’s rendering a chunk of
the tiles to the screen. AUDIENCE: But in this code, it’s not. COLTON OGDEN: In this code, no. The levels here are– it’s sufficiently
complicated to introduce. I mean, it’s not too
complicated to introduce. It’s pretty easy. But the consumption–
the processing here– is very light, because the
levels are fairly small. And even if we did have
really large levels, it’s sufficiently small to
not have to worry about it. But if we did get to a point where
your levels were 1,000 tiles or more, and then maybe those
tiles have additional, you just want to squeeze all
the performance possible out of your application. You could look into
just rendering a subset. It’s fairly simple to introduce
but just not something that we included in this assignment. Any other questions as to
how this sort of thing works? OK. So far, we’ve talked about
procedural level generation. We’ve talked about animation and
rendering and all that stuff. We haven’t really talked about
how to do tile collision. And we won’t go into a
terrible amount of detail, because the code is a little lengthy. It’ll be part of your assignment
to read over it and understand it, but it’s in the TileMap
class that we have. Basically, the whole gist
is that because we’re on a 2D tile array that’s fixed,
it’ll always be at 0, 0, at least in the model that we’ve
currently implemented. We can just convert
coordinates to tiles and then just check to see whether or not
the tiles at whatever that is are solid or not. Let’s say we wanted to look at the
top of our character in this case. So if we have our character here. For the sake of illustration, I
put him between two tiles above him just to show why we need to do
this the way that we are doing it. But you take the point here– his very top left, so
player.x and then player.y, which is effectively
their version of 0, 0. And then player.x plus
player.width minus– we do a minus one for
a lot of collisions so that he can walk between
blocks and stuff like that. Because if you don’t basically give
him slightly less than the amount– because he’s 16 pixels wide, and
the tiles are 16 pixels wide– if he’s between two blocks
and he wanted to fall down, he just won’t fall down, because
it’s still detecting a collision. Because if he’s on the hole here– let’s say this is the hole,
and these are the tiles here. The x plus the width– it’ll trigger a collision on
this tile and this tile still. So basically, you need to minimize
his collision box by one pixel to fit through 16 pixel gaps
essentially is what it boils down to. But the gist behind collision– in this case, this would
only apply when he’s jumping, because this is the
only time at which he can really collide with tiles that are above him. You would test for whatever
block falls on this pixel and whatever block falls on this pixel. And if either of them are
solid, you trigger collision. And if not, then there’s
no collision at all. So if he’s right here, for example– right directly beneath a tile–
it’s only going to check one tile. This point and this point are
both going to fall on this tile. But the reason that we want to
check for both points here and here is in the event that he is
beneath two separate tiles, because now this point’s
going to check this tile, and this one’s going to check this tile. We can’t just check this tile,
because if we only check this tile and there was no tile here
but there was a tile here, him jumping would still
not trigger a collision. It would think that it was
only looking here and not here. So for every collision on every
side we do of him effectively, we need to check both corners
of that edge effectively. So when he’s jumping,
we turn this point– this x, y– into a tile by
just dividing it by tile size. So we can say, player.x
divided by tile size plus one. That’s going to equal whatever
tile this is on the x. And then same thing for the y– we just divide the y by tile
size, and then we add 1 to it. And that will allow us
to get the exact tile. If we use those x, y that
we get from that operation, we get the exact tile at that y, x index
in our tile’s 2D array effectively. So we do that for jump. We check both corners
of the top of his head. We do the same thing for the
bottom, only at that time, we’re checking x, and
then y plus height, and then x plus width, y plus height. And then if we’re doing the
left edge, what are we checking? AUDIENCE: The bottom left and top left? COLTON OGDEN: We are. So that will be x0, y, and
then x0, y plus height. And then if it’s the
right edge, same thing. We check x plus width y, and then
we check x plus width y plus height. And so that’s the gist behind
collision detection in the distro here. And you can see it in
Mario if we go to TileMap. Point to tile– this is
effectively where it happens. On line 32, we’re basically returning– this bit of code here– 28 to 30– is a check. Because we can jump
over the map edge, we won’t be able to check at
tile y divided by TILE_SIZE plus 1, x divided by TILE_SIZE plus
one, because those will be nil. Those won’t exist,
because he’ll literally be outside the map boundaries. Same thing if he goes below it or he
goes beyond the left or right edge. So that’s all this code is here. It just makes sure that if we
do go beyond the map boundaries, we return nil. So that way, we can check nil rather
than getting a tile index error. And then on line 32 is the
operation that I just mentioned, which was we take the y– so this x and y that we pass in are
going to be the player’s actual x, y. When we pass those in, we’re just
going to get the tile at self.tiles, and then effectively y divided by
TILE_SIZE taken down to an integer, and then add 1. Because recall, tables are 1 indexed,
but the coordinates are 0 indexed. So this will result in a 0 indexed
outcome, so we want to add 1 to it. Same thing for here– math.floor(x)
divided by TILE_SIZE plus 1. So effectively, points to tiles. And then we’ll just
get a tile from that. And the tile– we can just check,
hey, is that tile solid or not? If it is, trigger collision. So that’s the gist behind being
able to do it in a platformer where everything is fixed. That’s sort of like a
shortcut we can take. Because now, what’s the
nice thing about this? What jumps out as being a super
nice thing about this algorithm, imagining that we have, let’s say,
10,000 tiles in our game world. So if you look and
see, all we’re doing is we’re just doing a simple
mathematical operation on what his x and y is, right? What’s the alternative to this? If we were doing this
via AABB, for example, we’d have to iterate over
every single tile, right? AUDIENCE: Can you summarize? To avoid iterating over
everything on the screen, you just check the column that
he’s in and the column tile? COLTON OGDEN: Yep. So the gist is he’s got an x and a y. The x and the y are going to be in
world coordinates, so his x could be 67 and his y could be 38
or something like that. They don’t map evenly to tiles. But if we divide those by whatever
the tile size is in our world– 16– that’s going to be the exact tile. We also have to add 1 to it, because
the tables in lua are 1 indexed. But we can index our
self.tiles at the x, y that we get from that–
the dividing by 16. And that will be the exact
tile that he’s colliding with. We don’t have to basically
have a collection of tiles that we iterate over
and check whether they collide with the player using
AABB collision detection like we’ve done before. Because recall, in Breakout,
we had the bricks, right? They all had their own x, y,
but they weren’t on a grid. They weren’t fixed. So we had to actually
take them and do an AABB. We had to iterate over them
and perform AABB on them, because there’s no deterministic way
to just index at them really quickly. It’s the same thing with
arrays versus linked lists. Because arrays– you can calculate
how far some value is given an index. You have instant access to it. It’s an order of one operation
as opposed to a linked list. If you want to try and
get to a particular value, you have to iterate through the
entire thing until you find it. AUDIENCE: Can you just look for the
column that you might be landing on? COLTON OGDEN: You’re getting the exact
tile at whatever your x divided by 16– or whatever your tile size is– and your y divided by 16 is. And you’re doing it, recall,
for two different points depending on what you’re looking for. If you’re looking for the tiles
that are above your character, you’re going to be
doing it for this point. So whatever this value
is– his base x, y– whatever that is divided by 16 and
then whatever that is divided by 16. And then that’ll get you whatever
tiles are directly above him. It will intersect with whatever
tile intersects with this point and whatever tile
intersects with this point. Same thing with here and here
if we’re looking on the left, here, here if we’re
looking on the bottom, and here and here if we’re
looking on the right side. And we check for collision
after he’s already moved so that these points will be
intersected with potential blocks. And that’s how we can check
whether it’s a collision or not. We do this when he moves and is
in some sort of movement state. AUDIENCE: So you’re still
doing collision detection with his actual
coordinates, but you’re just narrowing what you’re character width– COLTON OGDEN: Yep. we’re turning it from iterating
over every single tile to an instant operation, because
we can just mathematically get the exact tiles that
he’s at without having to worry about where he is in the map. It’s just instant access. And this only works because
we know the tiles are always fixed in the exact same locations. They’re always starting at 0, 0. They’re always going to be TILE_SIZE. Things get a little
more complicated when we introduce game objects, which
have their own independent x, y. And for those, you do
have to iterate over. You have basically a collection of game
objects or a collection of entities. Let’s say we have snails
in the game world. The snails aren’t going to be at
some fixed location every time. They can move continuously. So for those, we have to actually
keep them all in a container and then loop through them and say, has
my player collided with any of these? If he has, then trigger a
collision with that snail– kill it or kill the player if he’s
in a walking state or a jump state. And if he’s in a
falling state, then they should die, because he’s
colliding with them from the top. And you narrow down what collision you
check for, as you can see at the bottom here. Tile collision– when you’re
looking above your character, you’re only testing that when
you’re in the jumping state, because it’s the only time you need to. So that’s the only point at which you’ll
collide with tiles that are above you. When you’re in the falling state is
when you’ll check for tiles below you. And then you can interact
with tiles to your side when you’re in either the
jumping, falling, or moving state, so you should check for left and right
tiles in all three of those states. AUDIENCE: Shouldn’t you always test for
beneath you in case you get a chasm? COLTON OGDEN: In case what? AUDIENCE: In case you get a chasm. COLTON OGDEN: In case
you get chasm, yes. You’re correct. This should actually be
tested only when in the player falling state and player
walking state, yes. So the question was, shouldn’t you
be testing for tiles beneath you when you’re walking? And yes– not just falling,
but walking as well. This one only jumping, this
one falling and walking, and this one for jumping,
falling, and moving. Does that make sense– how we can take the x, y and
sort of turn that into a tile by just dividing it by 16? And do note the plus 1 as well,
because our tiles in our self.tiles are 1 indexed. And so when we divide
x, y by tile size, we’re going to get a 0 indexed coordinate. If our x is at 14, we’re
within the first tile. But if we divide that by
16, we’re going to get zero. So we need to add 1 to that so that
we get the first tile in the array still, which will be
whatever that tile is. So that’s how the collision works. It’s all implemented in TileMap here. And basically every state that the
player is in, which is in StatesEntity, and then player falling,
idle, jump, and walking– these are all states
that perform this check. They basically do all
the logic that’s here at the bottom, which is
testing in the player jumping state, falling state, and moving
state for left or right collision. And then in the falling state,
we check for collision below us. And then in jumping state, we
check for collision above us. That’s all done within
the states themselves. But the actual transformation
from pixels to tiles– that’s just a function
that we call from TileMap. It’s just a utility function. AUDIENCE: What’s the
function called again? COLTON OGDEN: It’s called pointToTile. So if you’re in TileMap on line 27– pointToTile(x, y). And the first little
bit here is just the bit that lets you basically
go outside the map bounds without getting a tile index error. So if it’s just outside the tile limits,
less than 0, or greater than width, just return nil. And so you can do a
check on nil to check to see whether TileMap
pointToTile is equal to nil or not when you do the collision. And if it is, then just
don’t do anything probably. But assuming that you’re within
the tile boundaries, on line 32 is where you do that transformation–
the math.floor, recall, because we want to get
integer values for these. We don’t want to get
fractional numbers, because you can’t index these tiles as fractional
numbers, although I’m not sure. I think you might be able to
in Lua generally– index a tile by a fractional number. But in this case, we just want integers. So we call math.floor on y
divided by TILE_SIZE plus 1, y divided by TILE_SIZE
then add 1 to that, and then we do the same thing for the x. So that’s the operation. And then wherever we want to check for
whatever tiles we want to collide for, we just call pointToTile on
those x and y coordinates. That’s the backbone behind all the
tile-based collision in the game effectively. Any questions as to how this works? Yes. AUDIENCE: So you’re only
detecting the collision of corners and not the edge itself? COLTON OGDEN: Correct, because you
don’t really need to check for the edge if you’re taking into consideration
the top and bottom corner, unless your entity is
sufficiently tall that they need to check for more than three tiles. In this case, our entity is
not more than two tiles tall, so we only need to check for
his top left, bottom left. If we’re doing a left collision,
top right, bottom right. If we’re doing a right collision, his
top left, top right for top and bottom left, bottom right for bottom. If you had an entity that
was eight tiles tall, you need to check every single
tile along his right side, which just means you need to iterate over
his entire height divided by tile size. And then just offset the y that you’re
checking for each of those tiles. Does that makes sense? OK, cool. All right. I alluded to this briefly
by mentioning state. I don’t know if I alluded so much to
the fact that we’re using entities. But in this distro, we’re introduced
to the concept of entities. An entity can be almost
anything you want it to be. In this distro, we’re
considering entities to basically be anything that’s
living or sentient moving around– in this case, the player or snails. Those are entities, and then
they just are subsets of entity. An entity is a very abstract thing. You’ll see it in a lot of game
engines and a lot of discussions about how to organize your
game and how to engineer it. Unity is probably the
most prominent adopter of what’s called the entity
component system, whereby you have everything in your game. Every single thing in
your game is an entity, and then every entity is
comprised of components. And these components
ultimately drive your behavior. It’s sort of like if you’re familiar
with composition over inheritance. If you’ve heard of this as a
software engineering thing, that’s effectively the same paradigm. Rather than inherit a bunch of
different things to be your– let’s say you have a base monster class. And then you have a goblin
that’s a subset of monster, so it inherits from monster. And then you have a goblin
warlord who inherits from goblin, and then you have an ancient goblin
warlord that inherits from that. Rather than have this
nested tree of inheritance, you adopt composition, which
means you take a base container, and then you fill it with
different components that represent what the behavior of your object is. So if you have an entity– let’s say you give it
a monster component. And then maybe you also give
it an ancient component, so it’s an ancient monster. And maybe you give it
a goblin component, so then it’s an ancient monster goblin. And then you give it
a warlord component, so it’s an ancient
goblin monster warlord. So it has all the
pieces that make it what it is without you having to create
this crazy chain of inheritance. That’s effectively what the model
of an entity component system is versus standard inheritance–
using that to drive the model of your problem. In this case, we’re not going
into crazy entity components. But I wanted to bring
it up, because Unity, which we’ll be covering in a few
weeks, is entirely component-based. Everything you write in
Unity is a component. And entities, whether they’re in
an entity component system or not, form the backbone of most large games. Most games that have
some complexity to them model most of the pieces within
them as entities that have behaviors and do things. And so in this case, entities
are snails and our player. And then separate from the tiles– when we do collision for that– we want to also check collision
on every entity with the player. So we make sure that the player’s
collided with the snail in this case, because that’s the only other
entities that they can be. But you can have an arbitrary
number of enemies if you want to. If you collide with an
entity– so just a for loop. So for entity in pairs of
entities, check collision. If you’re in the jump state, then die. If you’re in the fall
state, kill it, et cetera. When you’re doing most of your
entity to entity interaction stuff, that’s generally how you’ll model it. You’ll just iterate over everything
and then just collide everything. Depending on what
collides with what, you’ll just collide everything with everything
else and process interactions that way. That’s effectively how we do it. We have in the– I believe it’s in GameLevel. This maintains a reference
to a table full of entities, a table full of objects. Objects can be– we’ll talk
about that in a second– gems, and blocks, and bushes, and
stuff like that, and then a tile map. For every entity, we just update it. And then for every object, we update it. And then for every object in
objects, we render it as well. And then we render every entity. This is just sort of basic how
you would take a game world, populate it, and then
process and update it. Just containers, tables that maintain
a bunch of references to everything, and then just update them. The actual interaction
takes place in the– because they’re dependent
on what state we’re in. If you look at all the different states
for the player in the states slash entity folder, you’ll see, for example,
on line 62 of the player falling state, we’re iterating over every
object in the level.objects. And notice the player has
a reference to its level so that it can access
everything within it. And then within that level,
all the objects are stored. So all it needs to do is just
say, if the object collides the player and the object is
solid, then set our dy to zero, et cetera, et cetera. All this code’s actually
pretty easy to read through, so I would encourage
you to take a look at it and just understand how all
the collision and stuff is working between the player, the objects,
the blocks, and things like that– things like blocks are solid,
things like bushes are not solid. But that’s the gist. Have a collection of
objects or entities. And then depending on what state
you’re in, collide with some. And then depending on the
state, maybe that kills you, maybe that kills the enemy,
maybe nothing happens, maybe you become invincible. Maybe you collide with
a power-up game object, and that power-up triggers your
self.player.invincible is true. And then if
self.player.invincible is true, then maybe you render him
with a rainbow animation. And then in any of the functions where
he would collide and die with an enemy, he no longer dies, he just kills them. So that’s sort of the gist behind
how you would interact with objects and how to process it. Game objects are different. Like I said earlier, these are examples
of some of the objects here we can see. The gems on the bottom left
there are all in the distro. If you hit a block– and
if we have a few minutes, I’ll show you really
quickly how that works– if you hit a block, you’ll
have a chance to spawn a gem. If you collect the gen– which means
if you collide with that game object– increment your score 100. These are all other objects that I
didn’t have any time to implement. But just off the gate,
just as a mental exercise, how do you think we
could implement a ladder? Yeah. AUDIENCE: You would
just have a climb state. And if the player is touching a
ladder and presses a certain key, they would enter the climb state,
and that would cause them to go up. COLTON OGDEN: Correct. So what Tony said was if
they go onto a ladder, they should go into a climb state. And depending on whether
they’re in a climb state or not, if they press a button,
they should go up or down. And then you would check. If they’re at the top of the
ladder, get off the climb state, go into a walk state. Or if they’re at the bottom of
the ladder, go into a walk state. And that’s just another game
object that you just collide with, and then it’s a new state for you. Yes. AUDIENCE: You may actually
want it in a fall state, because that way you could have a ladder
that doesn’t actually go anywhere, it just gives you height. But you could use that
to jump over wide gaps. COLTON OGDEN: What
Tony said was you could have the ability to jump off a ladder. Is that what you Said Yeah. The ability to jump off a ladder so
that you can then use it as an obstacle. That’s absolutely true. Actually, in the mock up that we
saw up here, it’s super hard to see. I’ll see if I can maybe
zoom in on it here. Mario, Mario, graphics, and
then it’s called full sheet. The whole entire sheet that I used for
this lecture is called fullsheet.png. I don’t know what that is. So if you zoom in really high
here, we can see effectively what you were alluding to– right here, this little rope thing. I’m guessing for the sake
of this mock up that’s what they were trying to illustrate. But you have a game object that
lets you go into a climb state. Whether it’s a ladder
or whether it’s a rope, just add a new state for the player. If they’re in that climb state,
then we have this new animation which we saw in the sheet earlier,
which was their back or their front. And then they just climb
up it and just update it if they’re moving up or down the ladder. And then just give them
the ability to jump off. And then when you get to the
top or bottom, just get off. And you could think a lot of the same
thing with a lot of these obstacles, like the spikes here. If you’re jumping and you hit
it, you should probably die. And so you would check for if the
object.ID maybe is equal to spikes or whether object.lethal equals true. Same thing with this one. And then some obstacles are completely
cosmetic, like this mushroom here. In the case of the distro,
bushes and mushrooms and cacti and all those sorts of things
are just completely cosmetic, so you can walk through them. They don’t trigger collision, but
they’re rendered as game objects. They’re not part of the tile grid. They don’t get processed
in the same way as tiles. They’re not stored in the y, x. So that’s effectively how we
can start thinking about objects and how to give them behavior. Part of the assignment is
going to be adding a flag. So this flag is in the sprite sheet. So what you’ll do– and I’ll touch on this at
the end of the lecture here. We’re getting close to it. These keys here actually
at the bottom right– so part of the assignment will be to– it’s actually right here. So I’ll go over it really quickly. Ensure the player always
starts above solid land. So in this case, when James came up
here and ran, you ran the first example. The very first time that
we spawned the game, it generated a chasm right
where the player spawned at x1. And so he just falls to
his death if that happens. Just right off the gate,
anybody have any ideas as to what we could do to check
to see if we’re at solid land, assuming that the player’s
default start is at x1? AUDIENCE: At that tile,
check if it’s solid. If not, then just move it
there over x until it’s true. COLTON OGDEN: Yeah. What we probably want to do is
look all the way down the column, because we start towards the top. If we find that there’s no tiles
down there– it’s just pure chasm– we probably want to shift the player. And then random keys and locks– let me open up LevelMaker so we can
see what you’ll be interacting with, because most of what you’ll be
doing actually is in LevelMaker. It does a lot of what we did
before with just math.random, and then it will insert into objects. So objects is a table here. It will insert a game object
depending on some logic. So in this case, if we’re
generating a pillar, we have a chance to generate
a bush on the pillar. So if math.random(8) is 1, in this
case, we’re already generating a pillar. So we have an additional chance
that’s on top of the chance to generated a pillar– so basically, I think it’s a 1 in 64
chance on that particular iteration to generate a pillar with a bush. You just add a new
game object to objects. In this case, this is the
constructor for a game object. You give it an x, y, width,
height, and then a frame. And then the frame is relative to
whatever quad table matches the texture string here. So bushes is the texture, and
so whatever quad in bushes you want to give it– in this case, we just gave
it a random frame from that. And then a lot of the same
logic applies to other parts. This is another part where we
generate bushes just on flat land. We have a chance to generate a block– 1 in 10 chance this is a jump block. So here we have texture,
x, y, width, height, frame. Notice that we have
collidable is true, and this is how we can test to see whether
a tile is collidable or not. Hit is false, meaning that
we haven’t hit it yet. And if we have hit it, then we do this
code basically– onCollide gets called. You can see where this gets
called in the collision code for– if we look in Player. Player has check left collisions,
check right collisions, and check object collisions. It doesn’t have check
up and down collisions. There is a corner case
for both of those such that the logic had to be duplicated. I forget exactly why. But you basically get a list
of objects that you check for. Oh, the reason why is
because when you get the collided objects when
you’re in the jump state, you trigger the onCollide function. So let’s go to PlayerJumpState. If we’re in the jump state,
this is where we would basically check to see if we’ve gotten any
objects that collide with the player. If it’s solid, call its onCollide
function, object.onCollide, and then we just pass in the
object itself, basically. And so if you go back
to LevelMaker, that’s where we write the onCollide function. We write the onCollide function
within the game object here. So we just give it an
onCollide, remember, because functions are
first class citizens. We can just say onCollide
gets function obj, where obj is going to be this object. If it wasn’t hit already, one
in five chance to spawn a gem– so going to create a gem. It’s got all the same stuff in it. In this case, it has its own
function called onConsume. onConsume takes a player and an object. And then this is all
arbitrary, by the way. You can create whatever
functions you want. These are callback
functions, effectively. We’re just going to play the pickup
sound and then add 100 to our score. And then here, in the event
that we did get a gem, we tween it over the
course of 0.1 seconds. We tween its y to be from
below the block to up above, so it has an upwards
animation, effectively. And then we have another
sound that plays. But that’s effectively how
we’re spawning game objects. Game objects have textures,
x, y, width, height, and then you can give them callback
functions that you then execute wherever it’s relevant to you. In this case, you’ll only really
need to worry about onCollide, because the assignment is
create random keys and locks. They have to be the same color,
but you can choose them at random. If the player collides
with the key, then he should probably get some flag
that’s like key obtained is true or something like that. And then you go to the block
that spawns in the level, so you should spawn a
block with that same color. And then on collide, you should
unlock it, so get rid of the block and then spawn a new game object– the flag. And then that flag will
have its own on collide. And when you collide with
the flag, restart the level. And that’s effectively the
gist behind the problem set. So it probably shouldn’t take– I would say probably maybe 40 or 50
lines of code probably should do it. AUDIENCE: That game object– was that a class? It’s not a table. What is that? COLTON OGDEN: It is a class. There’s a GameObject class. A game object is basically– and I
realize I didn’t touch on it too much. In the context of this distro, you
could almost think of it as an entity. In this case, what I’ve
done is I’ve differentiated between living things
and non-living things as being entities versus game objects,
which is a semi-arbitrary distinction. But for a small project
like this, it makes sense. For a large project, I would probably
create everything as an entity and then give different
kinds of entities their own behavior and
their own components, sort of like how you do with
an entity component system. AUDIENCE: Are there two ways
to create a class in Lua, one with the curly brackets and
one with regular parentheses? COLTON OGDEN: There is, actually. So the question was, is there more
than one way to create a class in Lua whether it’s parentheses
or curly brackets? Yes. I don’t think I’ve ever
actually talked about this. Let’s go back to LevelMaker. If you instantiate a
class and that class takes in just a table as its only
argument, you can just pass in this. This effectively is that argument table. You don’t need parentheses. It’s effectively doing this– same thing, only you don’t
need the parentheses. AUDIENCE: And then that’s a
table that’s being passed in? COLTON OGDEN: Correct. It’s just an alternative
form of instantiation on things that only take a
table as their argument for when they get instantiated. AUDIENCE: And you can only
have one table in that case? COLTON OGDEN: Correct, yes. AUDIENCE: Wouldn’t it be easier
to create a new class which would have its own set
of game objects so you would create a gem which
would be helper dot gem, which would in turn create a game object? COLTON OGDEN: Can you
say that one more time? AUDIENCE: It’s kind of hard to explain. Wouldn’t it be easier to
create another class which would have your gem and everything,
and your gem in that class would be a game object? But in this class,
when you wanted a gem, you would say local gem
equals helper class dot gem. COLTON OGDEN: The
question was, wouldn’t it be easier to create a helper class that
would allow you to instantiate gems? Possibly. I think if you were going to design
this a little bit more robustly, and if this were going
to be a larger game, then you would just create a
subclass for blocks, gems, et cetera. To shrink the number of files that
we had in the distro and to sort of consolidate everything together
and put all of the level code together in one spot, I decided to just create
GameObject as an abstract class that you could then just create your
own behavior for within the actual constructor– which is this bit
here, which is just the table– and then allow you to override the
onCollide and onConsume functions. You can actually give it
whatever functions you want. You could give this some arbitrary named
function and then test for it later. This is almost like an
obscure way of inheritance. But I think if I were to engineer this
with the goal of making it a really large game, I would just subclass. I would just create a class for gem,
a class for block, a class for bush, et cetera, et cetera. It wasn’t strictly
necessary for this example, so we ended up keeping
everything a little bit more abstract in a sense– a
little bit more general purpose. But yeah, you could definitely
create classes for those. And if you were in an
entity component system, you could have a consumable component. And then that consumable
component would then allow you to give it some sort of
behavior that affects the player when the player consumes that object. In this case, a gem is a
consumable, so you would just give it a consumable component
with a texture of the gem and then give it a
callback function that just increments the player’s score. You could probably put that
in 10 or 15 lines of code. It would be pretty easy. And then blocks would
be a spawner component, so they have a chance to spawn. And then you would pass in that
spawner component a gem maybe, so it would have a chance to spawn the
gem that you passed into the spawner component– and then also a solid
component to say, oh, this is solid. So if I hit it, I should
trigger a collision and not be able to walk through it. So you just layer on these components. I would encourage you to think
about this way of composing your objects a little bit, particularly
as we get towards Unity, which makes a lot of use of this concept. In short, yes. I think that’s pretty much everything. Let me just go ahead. We’re running out of time here,
but like I said, one more time– make sure the player starts above
solid land, random color key and lock, and then make sure that when you get
the key, you can unlock the lock, and that spawns the goal. So this is all something you can
just add to the LevelMaker class, and it will all work with
your game level that way. And then you touch the goal
flag, then respawn the level. So today, we talked
about Super Mario Bros. The other big Nintendo
game of that era– arguably one of the greatest
of all time– is Zelda. So we’ll be talking about a very
simple Legend of Zelda game, where we just have a random
dungeon that we can go through, a top down perspective,
fight simple monsters, open chests, blow up
walls– that sort of thing. We’ll talk about triggers and events. And then we’ll talk about
hurt boxes, inventory, a very simple GUI for opening
up a menu, and then world states so that we can see which doors
have been blasted open so that they render appropriately and whatnot. And that’s it for Mario. Thanks a lot for coming. I’ll see you guys next time.

3 thoughts to “Super Mario Bros. – Lecture 4 – CS50’s Introduction to Game Development”

  1. OK, it's deceiving if you look at these as independent lectures. The four lectures given by this speaker, Colton Ogden (a faculty technologist at Harvard University), is part of a cs50 course on game development. You can find it here: https://online-learning.harvard.edu/course/cs50s-introduction-game-development. The course includes much more than just this Super Mario lecture. It's based on using the LÖVE2D game development engine and Lua programming language. You can audit these courses for free, but then you don't get a certificate of completion. I'm not sure how much weight those carry. I'd think the more important thing is what you learned and took away from the courses. Thanks to Harvard for allowing people to audit these courses for free.

Leave a Reply

Your email address will not be published. Required fields are marked *