Let me tell you, the vultures are terrifying. The thing the devs implemented with the shadow sweeping across the screen first is genius, since it acts as a warning and has you running for cover in terror since you know what's coming. You think the lizards the biggest threat until a vulture swoops down and snatches up a lizard like it's nothing. Makes for some really intense moments
This is basically a more daring alternative to the backflip for when a lizard is between you and where you want to go. The main difference is that where the backflip takes you over the head of a lizard chasing you from behind, this is useful for getting past a lizard you're running towards. The game needs more moves that aren't jumps, as basically all you've been able to do so far is jumping in different ways, and I hope this one can start to fill out that niche.
Then there are also a few new moves for the maze parts of the game. Originally Rain World was supposed to be 50/50 platforming and crawling in mazes, but we gravitated away from that as the platforming was simply more fun. Part of the reason for that was probably that in the mazes you had nothing else to do than move in 4 directions. The game will never go back to 50/50, but James wants to re-introduce a bit of maze crawling for one or two of the regions, and hopefully this new move set will make that more exciting. Testing in a room that's only narrow corridors with a bunch of lizards has yielded hopeful results - it's a fundamentally different game, with a lower tempo and more focused on sneaking and thinking ahead, but it can get pretty tense as well!
First up we have two vertical moves. Hitting up+jump will make the slugcat heave itself upwards quickly a few tiles, after which you face a bit of a slowdown penalty, which is supposed to make the average vertical speed somewhat break even. This is supposed to be useful in situations where you quickly want to get up and around a corner to get out of the field of vision of some creature. It is also useful when getting up to a ledge where you can switch to running or some other "non-crawling" means of locomotion, because when switching you can sort of bypass the penalty and get a little speed boost. The other move is the slugcat just stops pressing against the walls and drops down a chute. You trigger it by pressing down+jump, and the slugcat will continue falling for as long as you hold the down button. When adding NPC hearing (which should happen soon!) you should probably make an audible sound if you hit the bottom of the chute without grabbing on to the walls first, as a risk-reward balance for the speed gain you get.
Then we have a much requested feature - the ability to turn around in a narrow corridor. You do this by hitting the opposite direction button + jump, or up+jump if you're upside down in a vertical chute. I've intentionally made this move super awkward, as it is actually sort of too cramped for the slugcat to move around in there, and you shouldn't go butt-first into corridors to begin with >That being my main philosophy haha, I can still totally see how there are some situations where you're alone in a room with no lizards, and end up facing the wrong way in a corridor that goes on foreeeeeever, and it's just tedious. So here it is, at least it's possible now.
Last one is a horizontal "shuffle", triggered simply by pressing forward+jump. This one takes some skill though - if you just press jump at any place in a horizontal corridor, you get a pathetic little forward trust followed by a slowdown penalty that makes your net speed pretty much not break even. However, if you have solid terrain behind you as you hit the button, you get a good forward momentum with just a small penalty at the end. Basically the slugcat needs a solid block to push against. This one is especially useful when coming around a corner into a horizontal corridor
An insane amount of work went in to this, and I'm pretty much done, but probably not entirely done. We have folding wings, grabbing tentacles, etc etc etc ...
Interacting with player:
Hanging out:
Slow motion flying:
Cicada jumping:
And a bonus one, white lizard catching a cicada:
Here was the original sketch of the cicada
This stuff is all pretty much just logistics. The cool stuff is an "internal state" for each tracked creature that is kept inside the RelationshipTracker. An internal state is a mini data structure which keeps track of some relevant stuff. Let's make a simple example: I'm a cicada - I want to cause a slugcat trouble. Unless the slugcat is holding a spear, in which case I want to get away.
So the slugcat is on the ground, no spear, and its representation in my cicada head is comfortably within my AgressionTracker. Now the slugcat runs over and picks up the spear. Now the thing is that we're doing this thing where RW creature don't know what they can't know. So say that the spear and the slugcat are out of my sight as the weapon is equipped. Every frame I (the cicada) ask my RelationshipTracker what relationship I have to the slugcat. The RelationshipTracker can't access the data about whether or not the slugcat is holding a spear directly, it has to go ask the internal state representing the slugcat. The internal state has a bool which represents the slugcat holding a spear or not.
As I still think that the slugcat is unarmed, I'll continue to go after it. Until I come around a corner - then I suddenly have a visual on the slugcat, and the internal state is allowed to correct itself. The bool changes, and with this new information the nature of our inter-creature relationship shifts. The RelationshipTracker takes care of it, moves the slugcat to my ThreatTracker, and I can start scramming.
That's the theory, now let's take a look!
So notice how the cicada is harassing the player, until the player is armed, then it decides to get out of there. The player then goes underground and actually throws the spear and is not armed any more, but because the cicada can't see this it's state of fear is unchanged. The next time it saw the player it would reconsider. This is the sort of stuff that makes me all warm inside
I have this policy that I'll tell what I've done, not what I hope to do, but you can probably imagine the possibilities yourself. Social stuff such as befriending creatures, angering creatures, creatures reacting dynamically to different situations like trying to save their friends from the jaws of a lizard, etc.
Basically in the old set-up, I wrote the AI with a sort of limited idea about the sort of interactions I'd want to implement down the line. I though of creature relationships as static, one always eat the other, the other always flee. In that system it made sense to make it so that upon spotting another creature the first time, the AI takes a look at the relationship to that creature, and hands it off to a dedicated AI module to take care of it. If the relationship is "Eats" the creature gets sorted into the PreyTracker, if the relationship is "AfraidOf" it's dedicated to the ThreatTracker, etc.
This system is very limited, so what I've been doing today is writing a RelationshipTracker, which re-evaluates the relationships to other creatures frame-by-frame, and if the relationship changes it sorts it into the correct module. Sounds easy, but there's a lot going on. Basically each module lives its own life, deleting and shuffling the creatures it keeps track of as it wants. So I have to make sure that if a relationship is moved, it's properly removed where it was, properly added where it should be, and so on. The good part is that this has solved some other problems that were hanging around, such as creatures sometimes forgetting about threats they haven't seen in a while and then being unable to add them back, and the like.
The main tracker in a RW creature's AI is simply called the Tracker. This one basically keeps track of what creatures I know of, and where I think they might currently be (a notion that might be correct if I'm currently looking at them, and various degrees of incorrect if it was a while since I saw them). Each creature I know of has a representation in the Tracker, and this is sort of the UR representation for that creature, other modules refer to this one when thinking about a creature.
Other trackers such as PreyTracker and ThreatTracker has a reference to the creature representation in the Tracker, as well as some other module-specific data regarding that creature, such as how yummy it looks or whatever might be relevant.
The relationship tracker keeps goes through the representations in the Tracker, asks the AI what relationship it currently has to the creature in question, and if necessary moves stuff around to where they're supposed to be. For example, if I am only hunting the slugcat within a specific area, but the slugcat just exited that area, the slugcat representation needs to move from my PreyTracker to some other module.
The system seems to be working, the only thing that's needed now is to migrate it to the vulture and lizard as well. The other creatures use simpler AI's that won't really need this level of complexity.
So on to implementation! Past day or so was spent familiarizing myself with Joar's audio language and establishing a new workflow. Like the rest of my life right now, it is a horrific mess of text files:
But as hellish as it looks, I am super happy with the results so far! My gameplan was to first do an updated version of the Lingo SFX implementation for some practice with the language and to get something to work from. I had expected that to take at least until the weekend, but managed to get a basic sound-set that already sounds significantly better than that in just the afternoon! Super encouraging results.
Anyway, I hope that wasn't TOO boring. The actual audio work itself is literally days of me moving knobs around, so that would be even worse to read! I'll post some video of this rough audio build tomorrow when I have all the bat sfx triggers sorted out, and post updates on how that evolves through to a more final version.
Sooo, hanging from the ceiling... At first I thought I was going to be able to skip all the rope physics stuff, and just have it lose it's thread(tail? Spider web? Well, the dangly thing) if the thread hit terrain. However, it turned out that the rope hits terrain ALL the time, especially as these fellas are not wall climbers and therefor actually have to secure their ropes somewhere close to terrain. So, rope physics it is.
I'm not doing anything super fancy, just want something like the ninja rope in worms - basically whenever it hits a corner that corner becomes the new attachment coordinate until the previous attachment coordinate is visible again. Slowly but surely I'm getting there, through horrible hacks:
Oh, I'm not the dev. Just an avid fan.
Now you got me browsing the devlog again. It's really amazing how far the game has come
Also, mice dying
I know, right? They're so cute.
Starting to think about assets for the new dark region. This one we had to think quite a lot about, because the darkness has some particular qualities. Basically it's two conflicting factors. On one hand, the more little details, the better the shadows look. "Porous" tiles with lots of holes and stuff look good in the second layer, because you get to see the shadows move around in them which really displays the volume.
On the other hand, the darkness makes the level way harder to read. Mostly this is because the darkness breaks up the simple rule the rest of the game follows - dark = close, light = far away. Usually you can look at a room and assume that the dark parts are terrain you'll collide with, and the light parts are backdrop. The light sources in a dark room however are in the room - and in order for that to work visually I had to make it so that stuff in front of the light source were darkened so that they would look back-lit. This means that with the light source we don't just have a dark-light gradient, we have a dark-light-dark gradient.
This works in most cases, but it is admittedly not quite as clear as the ordinary palette. Also, it gets way worse the more little details there are, and the less solid the tiles. Soooo, the same factors that made the light look good.
My idea for a solution to this is that we'll be rolling with big, chunky blocks, but on the blocks there are protruding patterns that the shadows can play with. I think this will get both birds with one stone - because the general contours of the blocks are big and square you quickly get an idea of what's terrain and what's not, but if you look closer inside those big blocks of color there are smaller variations of shadow play.
Here are some sketches:
And here's a "final" (huge quotation marks there, everything always end up hundreds of kilometers away from the "final" sketches) concept:
Basically the area will consist of of big boxes, that are somehow bolted together with huge clamps. These clamps will run on the exterior of the boxes, and leave shadows. Then the boxes themselves will have a patterned surface, that will also get some shadows going. I'm currently mocking up some tiles, let's see how it goes!
Not sure what you mean by persitent. The AI is crazy complex, and the ecosystem will interact outside of your presence so it's common to enter a room and see lizards fighting for dominance or vultures hunting for food.
Here's two recent posts the devs wrote about the game's AI
Dynamic relationships! What is this thing I'm doing? Basically in the old set-up, I wrote the AI with a sort of limited idea about the sort of interactions I'd want to implement down the line. I though of creature relationships as static, one always eat the other, the other always flee. In that system it made sense to make it so that upon spotting another creature the first time, the AI takes a look at the relationship to that creature, and hands it off to a dedicated AI module to take care of it. If the relationship is "Eats" the creature gets sorted into the PreyTracker, if the relationship is "AfraidOf" it's dedicated to the ThreatTracker, etc.
This system is very limited, so what I've been doing today is writing a RelationshipTracker, which re-evaluates the relationships to other creatures frame-by-frame, and if the relationship changes it sorts it into the correct module. Sounds easy, but there's a lot going on. Basically each module lives its own life, deleting and shuffling the creatures it keeps track of as it wants. So I have to make sure that if a relationship is moved, it's properly removed where it was, properly added where it should be, and so on. The good part is that this has solved some other problems that were hanging around, such as creatures sometimes forgetting about threats they haven't seen in a while and then being unable to add them back, and the like.
The main tracker in a RW creature's AI is simply called the Tracker. This one basically keeps track of what creatures I know of, and where I think they might currently be (a notion that might be correct if I'm currently looking at them, and various degrees of incorrect if it was a while since I saw them). Each creature I know of has a representation in the Tracker, and this is sort of the UR representation for that creature, other modules refer to this one when thinking about a creature.
Other trackers such as PreyTracker and ThreatTracker has a reference to the creature representation in the Tracker, as well as some other module-specific data regarding that creature, such as how yummy it looks or whatever might be relevant.
The relationship tracker keeps goes through the representations in the Tracker, asks the AI what relationship it currently has to the creature in question, and if necessary moves stuff around to where they're supposed to be. For example, if I am only hunting the slugcat within a specific area, but the slugcat just exited that area, the slugcat representation needs to move from my PreyTracker to some other module.
This stuff is all pretty much just logistics. The cool stuff is an "internal state" for each tracked creature that is kept inside the RelationshipTracker. An internal state is a mini data structure which keeps track of some relevant stuff. Let's make a simple example: I'm a cicada - I want to cause a slugcat trouble. Unless the slugcat is holding a spear, in which case I want to get away.
So the slugcat is on the ground, no spear, and its representation in my cicada head is comfortably within my AgressionTracker. Now the slugcat runs over and picks up the spear. Now the thing is that we're doing this thing where RW creature don't know what they can't know. So say that the spear and the slugcat are out of my sight as the weapon is equipped. Every frame I (the cicada) ask my RelationshipTracker what relationship I have to the slugcat. The RelationshipTracker can't access the data about whether or not the slugcat is holding a spear directly, it has to go ask the internal state representing the slugcat. The internal state has a bool which represents the slugcat holding a spear or not.
As I still think that the slugcat is unarmed, I'll continue to go after it. Until I come around a corner - then I suddenly have a visual on the slugcat, and the internal state is allowed to correct itself. The bool changes, and with this new information the nature of our inter-creature relationship shifts. The RelationshipTracker takes care of it, moves the slugcat to my ThreatTracker, and I can start scramming.
That's the theory, now let's take a look!
So notice how the cicada is harassing the player, until the player is armed, then it decides to get out of there. The player then goes underground and actually throws the spear and is not armed any more, but because the cicada can't see this it's state of fear is unchanged. The next time it saw the player it would reconsider. This is the sort of stuff that makes me all warm inside
The system seems to be working, the only thing that's needed now is to migrate it to the vulture and lizard as well. The other creatures use simpler AI's that won't really need this level of complexity.
I have this policy that I'll tell what I've done, not what I hope to do, but you can probably imagine the possibilities yourself. Social stuff such as befriending creatures, angering creatures, creatures reacting dynamically to different situations like trying to save their friends from the jaws of a lizard, etc.
Working on vulture-lizard interaction. Here's a gif, there's quite a bit going on here so I'll do some explaining below ~
What the heatmap shows is the anticipated threat level for each tile in the lizard's threat tracker (note that I don't store this data, I generate it on the fly. So the framerate drops quite a bit with this visualization, but when the AI uses it it doesn't ask for every tile, just for relevant ones).
The lizard is inside the little box, which is open in both ends. This box is a sort of temporary shelter, because the vulture can reach in there with its neck - but it's juuust short of being able to reach the whole width of the box.
Basically the lizard merges two sets of data to get this map, the distance to the threat (naturally) and an Accessibility Map. See how the tiles close to terrain are bluer? And the tiles inside the box as well? That's one of the pre baked AI maps, the Accessibility Map, for the vulture. It allows the lizard to understand that inside the box is more safe than outside it. As you can see it's staying in the box, although nervously.
Also you might notice that all terrain has some bluer tiles around it - that's just because the vulture is big and close-to-terrain tiles are considered inaccessible to it for that reason. This might actually need fixing, because just because the vulture's center of mass can get close to a floor doesn't mean its neck is unable to.
The amount with which the Accessibility Map weights in on the final threat map scales with distance. You can see this particularly clearly when the vulture moves down towards the lower-right platform, the "safety" of the tiles surrounding the platform is removed.
This is the same effect that also finally drives the lizard out of the box. When the vulture does its peek-a-boo over at the left side, it's much closer and the safety of the box interior is disrupted. Now the danger gradient inside the box is steep enough for it to be worth it to dash for the other exit. So the lizard goes for it and heads for the room exit up in the corner instead.
You can also see how the lizard's perception of the vulture's position is only updated when there's visual contact, which is why the heatmap is static while the vulture is moving below the box. In actuality the map shouldn't be still during that time, it should move with a Ghost of the vulture that the lizard tracks. But right now the Tracker has some problems with creatures that can move on big 2D planes - it considers each tile 4-way branch and immediately spawns up ghosts to follow all of them, hits the ghost cap and makes the ghosts stop. So it becomes more of a simple "where did I last see the creature" kind of thing. I'll get around to it.
No problem! Been following the game for two years, it's been awesome seeing the AI and creatures developI want this game so bad, probably my most anticipated game next to Witcher 3, and a lot of that has to do with them both utilizing a complex ecology of creature AI that operates independent of the player's actions/location, though RW seems to be much more detailed in that regard. Thanks for continually posting the dev updates here Badass, always makes me more excited checking on this thread once in awhile to see the progress being made
Remember the final game is planned to be 12 regions
There's a reason why I disable GIFs on mobile GAF
