This PR replaces our current NPC AI with a [behavior tree
system](https://en.wikipedia.org/wiki/Behavior_tree_(artificial_intelligence,_robotics_and_control)).
Behavior trees are a common way of creating AI in which you place nodes
in a tree structure to define what actions an AI should take.
AI controllers defined a list of /datum/ai_planning_subtree types in
behavior_nodes. Each subtree was a self-contained unit that could call
queue_behavior() to fire off /datum/ai_behavior actions. The controller
iterated subtrees in order, each one deciding independently whether to
queue something and deciding whether the next subtree would run.
This has a few issues:
1. There's no real structure; you are just defining a list of things to
try in order.
2. There was a loooot of subtrees that were basically the same as
another but with some slight modification
3. It was hard to understand.
Controllers now define a single json file describing a tree of nodes.
The tree is composed of structural composites:
Sequence - do A, then B, then C (and so on)
Selector - try A, if it fails try B, then C (and so on)
Parallel - run A and B simultaneously, with configurable failure/success
policies and or looping behavior
Subplan - loop a child continiously
Along that we also have "Decorators". These are nodes that basically
check a condition (E.g.; do we have a combat target). These decorators
can be used to gate behavior and are re-useable across behavior trees.
They also have a concept known as "Observers". Which lets them cancel
lower priority behavior in case their condition changes (Which we check
whenever a signal fires that fits that specific decorator). This makes
the AI much more responsive to change in environment.
For behaviors, we still use the ai_behavior datums. These are the actual
behaviors such as "Move to X", "Attack X". The only major change is that
these can no longer sleep() since they now run in the ai_controller.
Lastly, we now also have subtrees, except now they are essentially
pieces of behavior tree that can be re-used, or even overriden at
runtime or as a variable. Allowing for making modular AI made out of
several smaller trees.
You can set variables on these nodes directly via the extension (see
below), which should reduce the need to make subtypes of behaviors by a
lot. All of these vars are saved on the JSON and will be applied at
runtime.
If you are using subtrees, you can also assign "bindings" to these
variables, which will allow instances of the subtree to override those
variables.
Since a tree structure with variables becomes hard to parse in a JSON,
I've made a VSCode extension to edit these JSONs:
https://marketplace.visualstudio.com/items?itemName=BehaviorTreeG.behaviortreeghttps://github.com/CabinetOnFire/BehaviorTreeG
<img width="1795" height="1268" alt="image"
src="https://github.com/user-attachments/assets/56aa2f0b-3cf9-449f-bca4-8281fca82db6"
/>
This extension allows you to edit the behavior tree JSONs, and browse
through all the behaviors/decorators/subtrees we have
If you'd like more info on how to build these AI check out the
learn_ai.md. I will also make a tutorial to go over more depth on what
the system offers because I kind of suck at doing technical write-ups.
Targetting has been changed to. I've made a new acquire_targets behavior
that takes a target_source (what am I targetting) and
targetting_strategy (what does the candidate need to fulfill to be
considered a target). This allows us to make composites targetting
combinations to reduce the amount of specific find_and_set esque
behaviors we had before. Not everything is ported to this system but
that would be a longer term goal.
I've added a new build_bt script that converts all the behavior tree
JSONs into compiled versions. Why is this needed? Because I wanted to
keep using defines in behavior trees, so we need a way to convert this
into literal values before we send it to DM. This script runs on compile
and should also run in CI (If I didn't fuck that up!). This saves to a
new build/ folder.
I've ported every single AI in the game to this system (except raptors,
Kobsa is working on those so should be in soon!), so I do expect some
bugs to come out of this. But I also fixed some issues that have
probably been in the game for a long time such as:
- Fixed penguins being unable to fish
- Fixed bileworms not being able to devour people
- Fixes goldgrubs not grubbing gold (they could not mine!)
- Lizards actually eat food they find
Either way, I'd reccomend a long TM on this.
1. (Hopefully) a better development experience for making AI
2. Less copy-paste for behaviors, we should be able to re-use more
pieces to make behavior
3. Behavior trees is a more common pattern in making AI, so it should be
easier to find resources to find out how to do things.
🆑 CabinetOnFire, Iamgoofball, SmartKar, Ben10omintrix
refactor: Replaces our AI system with behavior trees, porting all
datum/ai to it
/🆑
I will add this PR with more details down the line. I think I got the
big picture but its a big PR, so sorry if I missed something important.
---------
Co-authored-by: Iamgoofball <iamgoofball@gmail.com>
Co-authored-by: SmArtKar <44720187+SmArtKar@users.noreply.github.com>
Co-authored-by: Ghom <42542238+Ghommie@users.noreply.github.com>
Co-authored-by: Ben10Omintrix <138636438+Ben10Omintrix@users.noreply.github.com>
Co-authored-by: SyncIt21 <110812394+SyncIt21@users.noreply.github.com>
## About The Pull Request
This Pull Request adds a new logging system that uses a timeline to
track and visualize important events for specific datums.
This is done via a new window in which you can select a datum for
tracking, which adds it to the timeline. If this datum implements the
EVLOGGING macros, it can track important events onto this timeline. As
an example, we can log whenever an AI is deciding to make a new path, if
it decides to generate a new decisionmaking plan, it finishes an action,
or it decides to target someone/something.
We can select these events to see more information, and optionally get a
snapshot of important variables at the time this event was logged (like
the blackboard and current plan for AI controllers).
You can also filter out specific events / track info, which is done via
categories. Each event / piece of track info is given a category and if
you disable a category all events / track info in that category is
hidden. This lets you filter out things you might not care about.
<img width="2346" height="1209" alt="image"
src="https://github.com/user-attachments/assets/0763077c-e349-4c7c-b017-23d29e1d089b"
/>
_whoever thinks we didnt need advanced cleanbot logging is a noob_
In the video below I showcase how this works;
https://file.house/7nsOiqdvmSTxlsk3fs-e8g==.mp4
A cleanbot is roaming the halls, I turn on the event logger, click the
"pick target" button and click on the datum I'd like to track (the
cleanbot). This results in the cleanbot now tracking its events. I spawn
some dirt and the cleanbot decides to clean it, and I go through the
events; You can see theres different events being listed, such as when
the cleanbot starts targetting the dirt, when it cleans plan, when it
makes it JPS path and every time it moves over it.
The macros I've currently implemented are as follows:
**EVLOG_TEXT(DATUM, CATEGORY, INFO)**
Only adds text to the event logger window, no world-visuals
EVLOG_LOCATION(DATUM, CATEGORY, INFO, TURF)
Adds text to the event logger and adds an image to where that turf is.
EVLOG_TURFS(DATUM, CATEGORY, INFO, TURFS)
Adds text to the event logger and adds an image to each turf in the
TURFS list
EVLOG_LINES(DATUM, CATEGORY, INFO, TURF_A, TURF_B)
Adds text to the event logger and adds a line from turf_a to turf_B
EVLOG_PATH(DATUM, CATEGORY, INFO, TURFS)
Adds text to the event logger and visualizes a path from A to B (same
way as the pathfinding debugger, of which I moved the visualization
before to SSPathfinder)
In terms of performance, the logger is a singleton, and events are ONLY
logged if
1. The logger is running
2. The datum has the DF_EVLOGGING flag.
This means most of the time, logging an event is a single var lookup
(Since the runner is off by default). The DF_EVLOGGING flag is off by
default as well and has to be enabled by the event logger, or set
temporarily by a dev in code.
This system can easily be extended with more event types / visualization
types as well. (I'm thinking of datumizing the ones I have now)
The TGUI is still a bit of a mess, I would love some pointers because
I'm not really good at react so I just kind of hit it with a hammer
until it did what I wanted 😎
Also, all of this is based on VisLogging from Unreal Engine, so it will
have some likeness https://unreal-garden.com/tutorials/visual-logger/
## Why It's Good For The Game
This system allows us to debug more complex systems (like basic AI) in
an understandable and clear way. While the implementation cases are not
super common right now, extending this system could make debugging these
systems much more comprehensible, and hopefully lets more developers
help us with improving these systems. (plus, we LOVE timelines)
## Changelog
🆑 CabinetOnFire
refactor: Implements "Event Logging" an improved way for programmers to
debug specific datums.
/🆑
---------
Co-authored-by: Lucy <lucy@absolucy.moe>
## About The Pull Request
Bot huds would sometimes dissappear when they're pathing to anything
that arent beacons
## Why It's Good For The Game
Bot huds would sometimes dissappear when they're pathing to anything
that arent beacons
## Changelog
🆑
fix: fixes bot huds randomly dissappearing
/🆑
## About The Pull Request
Two main things.
Multiple instances of use of a proc in AI controllers seemingly assuming
the default behaviour will work for them, but what ends up happening is
`search_tactic` gets redefined and redefined with no defaut search range
parameter, so nothing ends up passed to the `search_tactic` child procs,
so they all call `oview` with `null` and this... somehow doesn't
runtime? Has behaviour that works some of the time??? I hate this
fucking language. Anyway.
Stoat steal items behaviour was completely broken and apparently was not
tested once since it was merged in. I've made the corrections I can, but
I haven't figured out why stoat AI never enters idle, so we have a
behaviour that leads to the stoat running up to an item, grabbing it,
and then just staying there, unmoving. I've sunk too many hours into
this, I'm just going to call it fixed and let someone else figure out
what exciting additions there need to be to a behaviour that was never
functional in the first place.
## Why It's Good For The Game
i don't know man i just want the pain to stop
okay, generally speaking, when people write AI behaviours, they want
those AI behaviours to do something and not just silently fail for six
months or longer
## Changelog
🆑
fix: Stoats have a chance to try and grab items like they always should
have.
/🆑
## About The Pull Request
this fixes some few sentient repairbot-related issues that were reported
a few days ago. ive also given sentient repairbots a new simple UI to
manage their materials.
while I was at it, i also fixed bots continuously trying to path to
unreachable beacons, which can be quite costly since they're regularly
very far away.
## Why It's Good For The Game
closes#88067
## Changelog
🆑
fix: repairbots no longer get flashed by their own welder
fix: repairbots no longer break glass tables they step on
fix: repairbots can no longer flush their own welders
fix: fixes some runtimes when emagged repairbots try to deconstruct
things
fix: fixes sentient repairbots not being able to see or remove their
material counts
/🆑
## About The Pull Request
adds repairbots to the game!
this pr serves as a massive rework and buff to floorbots. i was a bit
sad that they dont get built much anymore so ive given them tons of more
utilities and uses.
Repairbots still inherit to place tiles and repair breaches. but they
can now rebuild walls, rebuild windows and repair structure and
machinery. Also Ive given them voicelines to add more character to them.
In short, they are very depressed with their job (however they express
their happy go lucky attitude when u emag them where they will start
deconstructing the station)
to demonstrate capabilities, here's a slightly sped up clip of some
repairbots patching up an area that was maxcapped 4 times:
https://github.com/user-attachments/assets/bddac3b0-1984-4571-85d3-c5283dd7c0de
When repairbots feel threatened, they will retract into their little
toolbox which u can pick up and hold in ur hand, either to conveniently
carry and plop them down at breached sites, or to bash skull with it
U can build them using a toolbox, proximity sensor, cyborg arm and a
conveyor belt
## Why It's Good For The Game
refactors floorbots and makes them alot more useful tools for engineers
to use
## Changelog
🆑
refactor: floorbots have been refactored, please report any bugs
add: adds repairbots to the game!
/🆑
## About The Pull Request
mobs would disregard you for a bit if they went through a do_after, such
as goldgrub digging or medbots healing. this fixes that
## Why It's Good For The Game
fixes ai controllers resetting their targets post do_afters
## Changelog
🆑
fix: fixes ai controllers resetting their targets post do_afters
/🆑
## About The Pull Request
bot modes werent correctly reflecting the bot's AI states. this pr fixes
that. also fixes some issues where the bot's AI might get stuck
inbetween switching modes
## Why It's Good For The Game
closes#86771
## Changelog
🆑
fix: botkeeper now displays bot's correct states!
/🆑
## About The Pull Request
this is a revival of #82635 . i got permission from potato to reopen
this, he did almost all the work. i only just solved the conflicts and
fixed all the bugs that were preventing the original from being merged
(but it should be TMed first)
## Why It's Good For The Game
slightly improves the performance of basic mob AI
## Changelog
🆑
LemonInTheDark
refactor: able_to_run and incapacitated have been refactored to be event
based
/🆑
---------
Co-authored-by: LemonInTheDark <58055496+LemonInTheDark@users.noreply.github.com>
Co-authored-by: ZephyrTFA <matthew@tfaluc.com>
## About The Pull Request
this refactors honkbots into basic mobs. its mostly a faithful 1:1
refactor but i couldnt keep my hands to myselves so i gave them some new
behaviors.
honkbots now love playing with clowns, they will go seek out for clowns
and celebrate around them. also, if the honkbot finds a banana peel or a
slippery item near it, it will actively drag people onto them
honkbots will now go out of theirway to mess with secbots and annoy them
## Why It's Good For The Game
refactors hinkbots into basic bots and also undoes some of the silliness
i did in the previous basic bot prs. i also added lazylist support to
remove_thing_from_list.
## Changelog
🆑
refactor: honkbots are now basic mobs, please report any bugs
add: honkbots will try to slip people on banana peels
/🆑
## About The Pull Request
a recent change to silicon access meant all bots, simple and basic had
all access from roundstart
## Why It's Good For The Game
fixes bots having all access
Fixes https://github.com/tgstation/tgstation/issues/82671
## Changelog
🆑
fix: basic bots no longer start with all access
/🆑
## About The Pull Request
1. Simple bots use Manual Emotes like Basic Bots
2. Re-ups the CD for emoting to commission bots to 60 seconds, as it was
originally with Simple Bots
3. Punctuates the manual emotes
4. Deletes the fist bump message
## Why It's Good For The Game
1. Consistency.
2. These messages get a little... a lot spammy when walking around the
halls, especially if Robotics has made a few additional bots. The
original implementation had a CD of 60 seconds, which the AI behavior
halved. Upping it back up for consistency + less spam.
3. Looks a little less shoddy.
4. It's a cute message, but it's completely out of character for what
the bots are supposed to be "doing". They're "saluting" their
"commissioned officer", not meeting up with the bros. Also they don't
have fists! I understand saluting with their "robot appendage" but they
don't punch people!
## Changelog
🆑 Melbert
qol: Basic Bots will use their "salute" emotes less often.
del: Basic bot fists.
/🆑
## About The Pull Request
This PR does many things, I'll try to explain the basic/background stuff
to the main thing first:
1. Adds a new remote that allows a human to function like an AI. It
controls a fly that will fly around the station slowly, and when it
reaches a machine then the person can interact with it as if they were
an AI. This required changing a lot of silicon/AI checks with one that
also checks for this remote, and some messing with shared ui state.
2. Moves req_access from the obj and bot to ``/atom/movable`` which lets
it be shared between the two, no more copy-paste and one side lacking
features/checks/signals the other has.
3. Adds a check for AI config for AI-related station traits, which was
lacking prior
Now for the good part...
Adds a new station trait that replaces the AI with a Human.
This person is equipped with an AI headset (including Binary), an
advanced camera console, an omni door wand, the machine controller, and
their laws.
They are immune to the SAT's turrets (even if set to target borgs) and
are slow outside of the SAT, mimicing the actions of the AI.
They interact with the world through their advanced camera console,
which allows them to do most AI stuff needed, and the holopad they can
connect to without having to ring first (like Command can).
They are given a paper with the laws they must follow, but since they
are human they are able to bend it. Cyborgs that run the default lawset
are "slaved" to them via an unremovable law 0, so the Human AI can bend
the laws if they really need to (for their own survival n such), and
make the cyborgs obey their commands above laws, but in general this
shouldn't be a frequent occurrence. This does take into account the
unique AI trait, so it's not guaranteed Asimov.
When this station trait rolls, all Intellicards, AI uploads, and AI core
boards are destroyed and are unresearchable. They can be spawned by
admins in-game if necessary. Maybe in the future we can also exclude
Oldstation from this but I haven't really decided.
Extra perks:
Human AI spawns with a Robotic voicebox (unless they are a body purist)
and teleport blocking implant, so they can't use teleporters to bypass
their on-station slowdown.
They also have an infinite laser pointer that can be used to blind
through their camera console. This is unfortunately nerfed from the
recent borg balance PR that removed its stun. This was meant to be the
alternative to no longer being able to permanently lock borgs down like
AIs can (or more than one, for that matter).
They aren't affected by Roburgers, Acid, and Fuel's toxicity.
Bots salute them like they do Beepsky (which is now a trait)
They spawn with SyndEye to replace the AI's tracking ability
They do not have a bank account
### The machine remote
The machine remote has a little fly in it that flies to the machines it
is pointed to, working as the arms and legs of the Human AI. It scans
the machine and punches in the action the AI does, and is how the AI
accesses basically anything. This fly slowly moves from one machine to
the next, and can be recalled with Alt Click.
It works on machines and bots.
### Video (Low quality to fit Github)
https://github.com/tgstation/tgstation/assets/53777086/e16509f8-8bed-42b5-9fbf-7e37165a11e8
## Why It's Good For The Game
I've seen a funny screenshot one day of a person replacing the AI by
using a bunch of door remotes, camera console, crew monitoring console,
and a few other things. I've been thinking about that for a few years
and really wanted to make it official if not easier to make possible,
because it is an incredibly funny interaction.
This makes it a reality, and while they aren't as powerful as regular
AIs, I think it makes for better and funnier in-game moments. With the
same weight as Cargorilla (1), I hope this wouldn't be rolling too often
and ruin rounds, but instead show off the different capabilities that
Humans and AIs can do, to do the job of an AI. You win some you lose
some.
## Changelog
🆑 JohnFulpWillard, Tattax
add: Adds a new station trait job: The Human AI.
/🆑
---------
Co-authored-by: MrMelbert <51863163+MrMelbert@users.noreply.github.com>
## About The Pull Request
fixes the flaky error that would happen in cleanbot controllers. i
discovered thru the profiler that a proc i was using to stop tracking
unreachable targets and caching them to a blacklist in selectbehavior
had a bit of a high real time cost, so if the mob got deleted midway
through for any reason would cause a runtime error. ive moved this proc
off selectbehavior and im instead letting the move loop cache this
target into the blacklist if it reaches the max pathing attempts
## Why It's Good For The Game
fixes a runtime that sometimes happens in ai controllers
## Changelog
🆑
fix: fixes a runtime that sometimes happens in ai controllers
/🆑
## About The Pull Request
bots will automatically add things it cant reach to a temporary
blacklist and ignore them until the blacklist is cleared. this however
can cause issues if u move out of the way of a stationary medbot while
its healing u, so it deems u unreachable and not try to heal u until the
blacklist is cleared
## Why It's Good For The Game
bots will automatically add things it cant reach to a temporary
blacklist and ignore them until the blacklist is cleared. this however
can cause issues if u move out of the way of a stationary medbot while
its healing u, so it deems u unreachable and not try to heal u until the
blacklist is cleared
## Changelog
🆑
fix: stationary medbots will no longer ignore u for a while if u move
out of their way while healing
/🆑
## About The Pull Request
turns hygenic bots into basic bots. also now PAIs and people can play as
hygeinebots. and they can wear hats
## Why It's Good For The Game
transforms hyginebots into basic bots. their old AI used to handle all
the logic. i moved some of the logic to the mob itself so players can
also clean (or burn) things. also this pr will add pathing limits to
bots, in the case the jps movement thinks it can reach something, but
actually cant, in which case the bot will give up the chase
## Changelog
🆑
refactor: hygeinebots are now basic bots. please report all the bugs
fix: fixes hygenebots not being able to patrol
add: hygeinebots can now be controlled by Players
/🆑
## About The Pull Request
Consider this a continuation of
https://github.com/tgstation/tgstation/pull/80202
~~It seems I missed a few.~~
Edit: Modified per request to handle this more broadly. If a pawn gets
`qdel`'d, the ai controller should be set to off and get removed from
the list of active controllers, and all their actions should be
canceled.
Also adds some qdeleted checks to `finish_action()`, which can still run
after the pawn gets qdeleted as part of the `CancelActions()` chain.
## Why It's Good For The Game
Less spurious CI failures.
## Changelog
Nothing player facing really.
## About The Pull Request
i forgto to set this in my last prs closes#80204
also closes#80252
## Why It's Good For The Game
ais can summon bots
## Changelog
🆑
fix: AIs can summon bots again
fix: cleanbots can clean floors directly underneath them and prioritize
floors nearest to them
fix: medbots drop empty medkits now
/🆑
## About The Pull Request
<details><summary>A bunch of the numerous CI issues </summary>
</details>
You can view the full list of them here
https://github.com/Skyrat-SS13/Skyrat-tg/actions/runs/7148986054/job/19470671408.
What seems to be happening is, the `ai_controller` `fire()`s, and at
some point the the `pawn` var has become null from qdeletion. Many of
the `SelectBehaviors()` procs make use of that var, and then try to
access it without any safeties whatsoever.
I believe it is mainly happening because of long `do_after()`s and other
procs that sleep.
This PR just adds those safeties. I probably didn't get them all, but
this should fix the ones I have seen in CI. There may be a better
solution to cover all future cases of this but I will wait on feedback
to proceed. See below comments:
---
I don't know if you would rather this to always be checked at the
controller level instead (or in `able_to_plan()` perhaps?) but I could
do that if it's wanted. I wasn't sure if there were certain things that
depended on `SelectBehaviors()` running for cleanup so I opted against
that.
On that note, shouldn't we just be qdeleting the `ai_controller` when
the pawn gets qdeleted? Is that not already happening? And if not, is
there a reason for it? That would probably be the best way to handle
it...
## Why It's Good For The Game
I would like to stop seeing so many random CI failures, wouldn't you?
## Changelog
🆑
fix: fixes some AI runtimes that were caused by the pawn becoming null
/🆑
## About The Pull Request
this pr refacotrs cleanbots into basic bots. also adds a new skillchip
for janitrs. this skillchip will allow janitors to communicate with
cleanbots and order them around, like pointing at something and telling
them to clean it. also now the cleanbot has an inbuilt mop which it will
use to smack mice and cockroaches
## Why It's Good For The Game
refactors cleanbots into basic bots and fixes them getting stuck
sometimes while patrolling. also janitors being able to order them
around can make them a bit more useful as tools for the janitor
## Changelog
🆑
refactor: cleanbots are refactored into basic bots. please report all
bugs
fix: fixes cleanbots getting stuck sometimes while patrolling
add: janitors get a new skillchip which allow them to communicate with
cleanbots
/🆑
## About The Pull Request
refactors bots into basic bots. i decided to do medibots first because
they were the most complex bots and wanted to get them out of the way
first. if this pr gets merged then i will be rolling out the rest of the
bot refactors over the next months
medibots can now wear any type of hats! some of these hats can trigger
unique new recorded voicelines and interactions!
## Why It's Good For The Game
converts medbots into basic medbots. they are now a bit more responsive
than before but overall they should act similarly to how they did. also
adds the basic bot ai framework which has all the generic bot behaviors
that the next bots may or may not need to have in their ai. i tried my
best to improve their code and turned some of their bitflags into define
bitfields. this pr may need a careful review because i did it from
scratch as the old bot project branch was too outdated for me to use
## Changelog
🆑
refactor: medbots are now basic bots. please report any bugs
add: medbots can wear hats!
/🆑
