=Paper=
{{Paper
|id=Vol-3926/paper1
|storemode=property
|title=5000 Characters at Video Frame Rate using Declarative Programming
|pdfUrl=https://ceur-ws.org/Vol-3926/paper1.pdf
|volume=Vol-3926
|authors=Samuel Hill,Ian Horswill
|dblpUrl=https://dblp.org/rec/conf/exag/HillH24
}}
==5000 Characters at Video Frame Rate using Declarative Programming==
https://ceur-ws.org/Vol-3926/paper1.pdf
5000 Characters at Video Frame Rate using Declarative
Programming
Samuel Hill1, Ian Horswill1
1 Northwestern University, 2233 Tech Drive, Evanston, IL, 60208, USA
Abstract
Video games have historically been limited in the number of characters under AI control. This is all
the more true for games that use declarative systems which are generally used for turn-based or
otherwise non-real-time games. While a declarative system can obviously be used for a real-time
simulation, we wanted to test the limits of how far it can be scaled for large numbers of characters.
This paper shows how 5000 characters can run simultaneously and in real-time with 60fps
performance on a modern laptop using a declarative logic programming language.
Keywords
Declarative programming, Logic programming, Needs-based AI, crowd simulations 1
1. Introduction a planner, GOAP, that allows for declarative
definitions of various AIs in the GDBEdit database tool
Performance considerations have traditionally [11].
restricted the number of simultaneous NPCs, One of the earliest and most influential symbolic
particularly when those characters are driven by systems for games is the Inform 7 language [12], [13]
symbolic rules or other declarative methods. Many that allows designers to build interactive narrative
systems that do support this kind of character AI are systems using declarative statements. The Versu
also turn-based rather than real-time. In this paper, simulationist narrative system [14], [15] used a
we show that 5000 The Sims-style characters can be custom logic programming language, Praxis, based on
run simultaneously on a modern laptop using a an exotic modal logic called eremic logic (aka
version of logic programming. exclusion logic) [16]. The Lume system [17] made
Although Sims-style ‘needs-based AI’ is relatively extensive use of Prolog’s definite clause grammars
simple, and the rendering of the characters in our [18], [19] for text generation. Lapeyrade has also used
demo is extremely simple (particle-like dots on the Prolog for better character decision making [20].
screen), this still demonstrates that symbolic, All these games and systems use declarative
declarative techniques can be pushed well beyond programming for either some component of character
what would have been though possible. In this paper, control and/or social reasoning. Some are turn-based
we will discuss the system, and the techniques used to games like in the Versu platform, others are batch like
make it so efficient. jobs used in Townlikes (that still need to run fast but
not at any particular frame rate), and others are real-
2. Related Work time systems like with GOAP in F.E.A.R.. Even when
used for real-time character control the number of
Many AI-based games have used symbolic declarative
NPCs is usually relatively low and the portion of the
systems - logic, symbolic rules, or some kind of rule
game logic that is declarative are the relations
engine - for social reasoning and character control.
between possible actions and the world state. The
Façade [1] used a reactive planner, ABL [2], that
code for how a Goal executes is still usually written in
incorporated a forward-chaining production system.
an imperative manner [11].
Prom Week [3] also used a forward-chaining
There are also entire games that have been built
production system, Comme Il Faut [4]. CiF’s successor
using declarative languages, such as those built with
the Ensemble Engine [5] is similarly a forward-
the video game description language - VGDL [21].
chaining production system. The Sims 3 [6] used a
Currently this technique is used to generate 2D tile
rule system to script the interactions between
arcade style games with the help of ASP and
situations, personality traits, and actions available to
evolutionary search [22], [23]. These games are
a given character [7]. MKULTRA [8] and City of
playable by a human and are fully declarative -
Gangsters [9] both used logic programming to
everything from character controls to game logic to
implement social reasoning [10]. F.E.A.R. makes use of
the tilemap itself are declared in text files that are
11th Experimental Artificial Intelligence in Games Workshop, Copyright © 2024 for this paper by its authors. Use permitted under
Creative Commons License Attribution 4.0 International (CC BY 4.0).
November 19, 2024, Lexington, Kentucky, USA.
CEUR
ceur-ws.org
Workshop ISSN 1613-0073
Proceedings
used to generate a game. The purpose of these games have been selected/assigned to them, potentially
is largely to allow for testing of general video game getting rewarded for the various tasks. When no more
playing (GVGP) algorithms on several different games actions remain in the queue of selected actions the
[24]. NPC searches for all the advertisements on all the
In large part because of the expense of AI objects around them then scores each action based on
simulation, games involving large-scale character their needs and selects the best scoring
simulation are relatively rare. The best known is advertisement. The equation for selecting the best
Dwarf Fortress [25], which supports real-time scoring advertisement is 𝑎𝑟𝑔𝑚𝑎𝑥(𝑆) where 𝑆 is the
simulation of small hundreds of characters. Achieving set of all scored advertisements for a given character.
this level of performance requires implementation in The complexity of scoring all advertisements is
C++ and significant programmer effort to optimize 𝑂(𝑁𝐿𝑀), where 𝑁 is the number of people, 𝐿 is the
cache locality and minimize pointer chasing. objects nearby, and 𝑀 is the advertisements at each
RimWorld is a very similar game that also involves location. Once you have scored all advertisements for
social simulation for the purpose of storytelling [26]. all characters, you can run the argmax equation for
Caves Of Qud is a game that is more like adventure each character. Since argmax is operating on the set of
mode in DF and, similarly, features hundreds of all advertisements for each person, the complexity of
characters that all have various social relations and selecting an action from the scored set is 𝑂(𝑁𝐿𝑀),
need to choose what action to take each tick [27]. where 𝑁 is the number of people, 𝐿 is the objects
Unlike DF Fortress mode and RimWorld, this rogue- nearby, and 𝑀 is the advertisements. Given that these
like experience is turn based and as such doesn’t need two steps are each of the same complexity, and that
to run at a particular frame rate. While all these games the two iterations are sequential, the total complexity
feature large numbers of characters acting in a world, of action selection is 𝑂(𝑁𝐿𝑀).
none use declarative systems for the control of these The selected advertised action sequence is then
NPCs. added to the queue of actions to be completed. Needs-
Some large-scale crowd simulation exist for use in based AI has always had the ability to sequence
film and other rendered media forms [28], [29]. Some actions, a technique called action chaining [34], but in
of these tools even allow the behaviors of the agents The Sims 4 the developers wanted to be able to have
to be stated in sentence form for specific decision more interesting action sequences. To accomplish this
nodes [30]. These tools target rendered scenes for the developers allowed the system to simultaneously
film and television and as such do not need to run in dispatch multiple actions from the queue when those
real-time. actions would not interfere with one another [35].
Some crowd simulation techniques have been The Sims 3 had a goal of creating a larger varied
used to study crowd dynamics and the creation of living world, and to do this while remaining playable
autonomous agents for visualization [31], [32]. Shao several optimizations were employed [36]. Some
have claimed real-time performance at 30fps when execution optimizations such as hierarchical planning
animating up to 1400 characters moving and and commodity-interaction maps as well as some
pathfinding around a 3D scene on 2007 hardware level of detail (LoD) optimizations such as auto-satisfy
[31]. This was done with over 50,000 lines of C++ and curves and story progression [36]. Hierarchical
uses a simple action selection method reminiscent of planning takes the usually 𝑂(𝑁𝐿𝑀) approach of
needs-based AI. searching for every person (𝑁), for every location (𝐿),
and then for every advertised action (𝑀), and makes
3. Needs-based AI it 𝑂(𝑁 + 𝐿 + 𝑀) by instead selecting each step one at
a time. Commodity-interaction maps are essentially
Needs-based AI is a famous technique for controlling
dictionaries of needs and the actions that satisfy those
characters in video games that was created for The
needs, and they are used to trade off compute time
Sims [33], [34], and has been used in other games like
searching over all advertisements on all objects to see
Roller Coaster Kingdom and an RPG for Lord Of the
if they satisfy a sim’s needs with storage cost. The LoD
Rings [34].
optimizations tick the actions of sims not in the
Needs-based AI attempts to fulfill various needs
current focused level when they come into focus
for NPCs by searching over the actions of each object
(auto-satisfy), and tick certain subsystems like the
available to them (i.e. advertisements) and weighting
wants of the town itself at a lower rate (story
the cost of completing the action against each
progression). As a result, although the simulation
character’s needs with some scoring function [34].
notionally supports hundreds of characters, the vast
Each character goes about completing actions that
majority of them are effectively idle at any given time.
Figure 1: State Fair Sim running in Unity
As well, although The Sims 3 has many more possible degree to which NPCs attenuate distance. The value
motives (needs) than the 8 used in previous versions, 2870 was not chosen at random, it relates to the size
any individual character has a small number of those of the map that the characters move on and effectively
possible motives. scales 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 2 to be between 1 and 30 (as opposed
to a max of 82369).
3.1. Scoring functions One more simplification in the current demo is
The scoring function that we use here is the most that all locations only advertise a single action. In
sophisticated function listed in Zubeks description of effect, a location has a type that correlates to a need.
As such, the complexity of this scoring algorithm can
needs-based AI, an attenuated need-delta scoring
be given as 𝑂(𝑁𝐿), where 𝑁 is the number of NPCs
function that is additionally attenuated based on
distance [34]. One difference in our version is that that need to select actions and 𝐿 is the number of
locations to search over.
each action can only satisfy one need. In the version of
action performance outlined by Zubek, some actions
can be sequences, and this can potentially allow for 4. Real-time logic programming
multiple needs to be fulfilled by an advertisement. As To test the scaling limits of declarative symbolic
such, the algorithm for calculating an advertised logical methods for real-time NPC control, we built a
action score must sum the attenuated scoring across state-fair simulator, as one might find in a tycoon
all needs. Instead of summing across all needs, with game, using a logic-programming implementation of
each action only satisfying one need, we can simplify needs-based AI. NPC’s wander through the state fair
the attenuated need-delta scoring with distance eating, drinking, seeking entertainments, and so on.
attenuation to: The simulation consists of a map with 367 locations,
onto which we spawn a few thousand NPCs. Each
𝑠𝑐𝑜𝑟𝑒 = 𝐷(𝐴𝑛𝑒𝑒𝑑 (𝜈𝑐𝑢𝑟𝑟𝑒𝑛𝑡 ) − 𝐴𝑛𝑒𝑒𝑑 (𝜈𝑐𝑢𝑟𝑟𝑒𝑛𝑡 + 𝛿)) character has 6 motivations (hunger, bathroom,
games, amusement, music, and shopping) that
where 𝜈𝑐𝑢𝑟𝑟𝑒𝑛𝑡 is the current need value of a particular influence their action selection.
need type, 𝛿 advertised need delta, 𝐴𝑛𝑒𝑒𝑑 is an
attenuation function, and 𝐷 is the distance 4.1. Needs-based AI in logic
10
attenuation function. 𝐴𝑛𝑒𝑒𝑑 (𝑥) = as is discussed in
𝑥
programming
𝑥
Zubek, but instead of the mentioned 𝐷(𝑥) = The logic for the core steps of scoring all available
𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒 2
𝑥
function ours is 𝐷(𝑥) = 𝑑𝑖𝑠𝑡𝑎𝑛𝑐𝑒2
to tune down the actions and then selecting the best action for each
1+( )
2870
character is contained in the few lines of code found in language implementation are outside the scope of this
Code Fragment 1 and 2. For a discussion of the paper but see [37], [38] for more information.
programming language used, see [37], [38], but the Another important optimization might be called
following fragments can be explained thusly: implicit need decay. Normally, a needs-based AI
Code Fragment 1 defines a predicate that is true system would update each need state of each
when a given person performing a given advertised character on every tick. For our system, this would be
action at a given location has a given score. It states prohibitively expensive. Instead, we store
that it has that score when the location advertises that timestamped need values; a need value is represented
action, it fulfills some need, and score is the result of as a tuple of (𝑣, 𝑡, 𝜌) stating that the need had value 𝜈
distance-based scoring, above, along with some noise at time 𝑡, and was decaying at rate 𝜌. We can then
to add a small amount of randomness to the action compute the current need value on demand as:
selection:
𝜈𝑐𝑢𝑟𝑟𝑒𝑛𝑡 = max(𝜈 − (𝑡 − 𝑡𝑛𝑜𝑤 ) ∗ 𝜌)
Code Fragment 1
Action Scoring with 𝑡𝑛𝑜𝑤 being the current time. Since these values
PersonActionScoring = Predicate("…", are read infrequently but would otherwise be updated
person.Indexed, actionType.Indexed, frequently, this is much more efficient. As is
location.Indexed, score).If(
recommended by Zubek, decay rates for a given need
ReadyToSelectAction,
CoordForPerson[person, coords], vary slightly from character to character to give them
LocationAdvertisements, slightly different “personalities”. These decay rates
deltaOffset == delta + are chosen randomly (Appendix A, Code Fragment 3).
RandomFloat[-2, 2],
ActionSatisfiesNeed, 4.3. Action States
FairgroundLocations[location, __,
otherCoords], For characters to perform a selected action they must
NeedByType[person, needType, need], move to the location of the selected action and then
DistanceScoring[need, deltaOffset, take some steps to complete the action before the
coords, otherCoords, need is satisfied. While we could allow for completion
score]); of actions instantaneously, to be more in line with The
Sims implementation - as specified by Zubek – we
Given this, Code Fragment 2 defines a predicate instead have some action performance before a need
stating a given action/location pair is best for a given is satisfied.
character (note that this is essentially the same as the When an action is initiated, the system tells the
argmax equation, just written in logic): unity code to start moving the character, and then
marks the character as being in progress towards that
Code Fragment 2 location. When the character arrives at the location
Action Selection
they are marked as no longer moving towards their
PersonActionBestScore = Predicate("…", destination. When the action is completed, unity
person.Key, actionType.Indexed,
informs the system, and the system updates the need
location.Indexed).If(
ReadyToSelectAction[tempPerson], states, and generates a new next action (Appendix A,
Maximal((person, actionType, Code Fragment 6, 7, 8, 11). In place of the animations
location), score, that would normally be running when performing an
PersonActionScoring[tempPerson, action, we simulate an animation running by starting
actionType, location, score] & a timer that indicates the selected action is being done
person == tempPerson)); (Appendix A, Code Fragment 9).
Unlike in The Sims all actions in this simulation
are individual, there is no chaining of actions nor are
4.2. Implementation optimizations there any concurrent interactions. This simplification
The performance of the system is dependent on is what allows us to use a timer for each person taking
several optimizations. Many of these have to do with an action, when the timer is running the “animation”
the specific programming language implementation: is playing and when the timer completes the character
strong typing, bottom-up execution, parallel
execution, native code compilation. The details of the
Figure 2: Performance of the State Fair Sim
has completed the action and finished its animation. 5. Performance
While we do not have proper animations for each sim,
when on a timer for an action the sim is moving With 5000 NPCs running this simulation on an Apple
randomly around the location that they have selected M3 Pro we are able to achieve 60fps or better for 95%
an action at. of frames when running some performance tests. Our
average inner-frame time was 13.87ms with a
4.4. Destinations & Unity Interface standard deviation of 2.077ms. Figure 3 shows our
performance curve once stable (from 12500 to 25000
For control of character movements, we have a simple
ticks) which is somewhat bell curve shaped with a
interface between the action selection simulation and
longer tail to the right. Most of the tail to the right is
the Unity code that performs movements. The
likely garbage collection or some other Unity process
Simulog code provides a table of people and the
as that performance tail is not present in Figure 4
destination they are moving to (Appendix A, Code
when we show the performance curve for just the rule
Fragment 7) that the Unity code iterates over each tick
execution times.
to adjust NPC positions. The Unity code in turn
provides a function that we can call to ask if someone
has arrived at their destination. With this simple
interface of action selection assigning destinations to
the Unity code and Unity only needing to report back
when someone has arrived at the destination, we are
able to change out the underlying movement
implementation without needing to do any
modifications to the action selection algorithm.
Currently, the movement is all simple vector addition
moving sims at a set speed straight towards their
destination.
simulation on 5000 characters without any action
selection takes about 1ms.
Figure 3: State Fair Sim Inter-Frame Intervals Figure 5: Performance of Action Selection
After these baseline costs, the variation in
performance largely comes down to the number of
characters selecting actions on any given tick. The
performance cost for one NPC selecting an action
averages to just under 0.3ms. This gap between each
number of agents selecting is clearly visible in Figure
2 and 5 with both the color mapped data points and
green lines indicating the moving average for each
tier. The same gaps are not visible in the inter-frame
time data (grey dots in Figure 2), although there is
clearly a correlation between the range of
performance for rule execution and for total inter-
frame time.
Figure 4: Action Selection Rule Execution Times
5.1. Pushing character counts
The graph in Figure 2 shows the progression of the
simulation over a 25,000-tick run time. The first 5000 The performance shown above is only possible when
ticks spawn one person every tick causing the slow we first compile our rules down to native C# code, the
rise in compute time for both the action selection process for which is described in [37], [38], and then
mechanism and the inner-frame updates. For the next we run our compiled code in parallel. The type of
7500 ticks characters begin to arrive at their first parallelism employed is per predicate, not per sim,
destination and need to select their next action, utilizing the dependency graph for all predicates to
creating a tiered slope in the action selection spin up tasks that run the compiled rules for a given
performance as various numbers of characters need predicate only when its dependencies have finished.
to select an action. From 12500 ticks on we have a Without these optimizations the maximum character
steady state where performance is consistent and count at nearly the same performance is 2000 NPCs.
linear with NPCs now moving around and selecting One optimization that could be employed to
actions at a rate of 13.7 actions selected on average further stabilize performance (and potentially allow
per tick for our sample with a σ of 3.8 actions. for more NPCs) is to throttle the number of characters
The cost of moving 5000 NPCs and rendering the that are selecting actions on a given tick. With a limit
scene (amongst other functions that Unity runs) is the on the number of characters selecting actions that is
difference between the action selection rule execution slightly higher than the un-limited average number
time and the inner-frame time. As can be seen in selecting you could effectively cap the cost of doing
Figure 1, this difference (looking from 5000 ticks to action selection while still eventually performing the
9000 ticks where action selection has not yet taken same sets of needs-based calculations. This technique
off) is about 7ms. There is also a baseline cost to was not employed in this demo but a commercial
running the rest of the declarative simulation when system that cares more about never dropping below
not performing action selection. As can be seen in 60fps may want to implement such a throttle.
Figures 2 and 4, the cost of running our declarative
6. Conclusion [15] R. P. Evans and E. Short, “The AI Architecture of
Versu”.
AI character control based on declarative [16] R. Evans, “Introducing Exclusion Logic as a
programming can be surprisingly performant. Deontic Logic,” in Deontic Logic in Computer
Through a combination of compilation, parallel Science, vol. 6181, G. Governatori and G. Sartor,
evaluation, and careful optimization, thousands of Eds., in Lecture Notes in Computer Science, vol.
characters can be run at video frame rate. Although 6181. , Berlin, Heidelberg: Springer Berlin
Heidelberg, 2010, pp. 179–195. doi:
this demonstration involves simple needs-based AI,
10.1007/978-3-642-14183-6_14.
we intend to investigate more complicated techniques [17] S. Mason, C. Stagg, and N. Wardrip-Fruin, “Lume:
in the future. a system for procedural story generation,” in
Proceedings of the 14th International Conference
References on the Foundations of Digital Games, San Luis
Obispo California USA: ACM, Aug. 2019, pp. 1–9.
[1] M. Mateas and A. Stern, Façade. (2005). doi: 10.1145/3337722.3337759.
[2] M. Mateas and A. Stern, “A behavior language for [18] F. C. N. Pereira and D. H. D. Warren, “Definite
story-based believable agents,” IEEE Intell. Syst., clause grammars for language analysis—A
vol. 17, no. 4, pp. 39–47, Jul. 2002, doi: survey of the formalism and a comparison with
10.1109/MIS.2002.1024751. augmented transition networks,” Artif. Intell., vol.
[3] J. McCoy, M. Treanor, B. Samuel, A. A. Reed, N. 13, no. 3, pp. 231–278, May 1980, doi:
Wardrip-Fruin, and M. Mateas, “Prom week,” in 10.1016/0004-3702(80)90003-X.
Proceedings of the International Conference on [19] F. Pereira and S. M. Shieber, “Prolog and Natural-
the Foundations of Digital Games, in FDG ’12. New Language Analysis,” 1987. [Online]. Available:
York, NY, USA: Association for Computing https://api.semanticscholar.org/CorpusID:2642
Machinery, May 2012, pp. 235–237. doi: 03475
10.1145/2282338.2282384. [20] S. Lapeyrade, “Reasoning with Ontologies for
[4] J. McCoy, M. Treanor, B. Samuel, N. Wardrip- Non-player Character’s Decision-Making in
Fruin, and M. Mateas, “Comme il Faut: A System Games,” Proc. AAAI Conf. Artif. Intell. Interact.
for Authoring Playable Social Models,” Proc. AAAI Digit. Entertain., vol. 18, no. 1, Art. no. 1, Oct.
Conf. Artif. Intell. Interact. Digit. Entertain., vol. 7, 2022, doi: 10.1609/aiide.v18i1.21980.
no. 1, pp. 158–163, Oct. 2011, doi: [21] T. Schaul, “A video game description language for
10.1609/aiide.v7i1.12454. model-based or interactive learning,” in 2013
[5] B. Samuel, A. A. Reed, P. Maddaloni, M. Mateas, IEEE Conference on Computational Inteligence in
and N. Wardrip-Fruin, “The Ensemble Engine: Games (CIG), Niagara Falls, ON, Canada: IEEE,
Next-Generation Social Physics”. Aug. 2013, pp. 1–8. doi:
[6] The Sims 3. (2009). Maxis. 10.1109/CIG.2013.6633610.
[7] R. Evans, “AI challenges in Sims 3,” Artif. Intell. [22] T. S. Nielsen, G. A. B. Barros, J. Togelius, and M. J.
Interact. Digit. Entertain., 2009. Nelson, “Towards generating arcade game rules
[8] I. Horswill, “Postmortem: MKULTRA, An with VGDL,” in 2015 IEEE Conference on
Experimental AI-Based Game,” Proc. AAAI Conf. Computational Intelligence and Games (CIG),
Artif. Intell. Interact. Digit. Entertain., vol. 14, no. Tainan: IEEE, Aug. 2015, pp. 185–192. doi:
1, Art. no. 1, Sep. 2018, doi: 10.1109/CIG.2015.7317941.
10.1609/aiide.v14i1.13027. [23] M. J. Nelson and M. Mateas, “Towards Automated
[9] City of Gangsters. (2021). SomaSim, Chicago. Game Design,” in AI*IA 2007: Artificial
[10] R. Zubek, I. Horswill, E. Robison, and M. Viglione, Intelligence and Human-Oriented Computing, vol.
“Social Modeling via Logic Programming in City 4733, R. Basili and M. T. Pazienza, Eds., in Lecture
of Gangsters,” Proc. AAAI Conf. Artif. Intell. Notes in Computer Science, vol. 4733. , Berlin,
Interact. Digit. Entertain., vol. 17, no. 1, pp. 220– Heidelberg: Springer Berlin Heidelberg, 2007,
226, Oct. 2021, doi: 10.1609/aiide.v17i1.18912. pp. 626–637. doi: 10.1007/978-3-540-74782-
[11] J. Orkin, “Three States and a Plan: The A.I. of 6_54.
F.E.A.R.,” 2006. [24] J. Levine et al., “General Video Game Playing,”
[12] G. Nelson, Inform 7. (2006). Schloss Dagstuhl – Leibniz-Zentrum für
[13] G. Nelson, “NATURAL LANGUAGE, SEMANTIC Informatik, 2013, p. 7 pages, 336517 bytes. doi:
ANALYSIS AND INTERACTIVE FICTION,” 2006. 10.4230/DFU.VOL6.12191.77.
[14] R. Evans and E. Short, “Versu—A Simulationist [25] T. Adams and Z. Adams, Slaves to Armok: God of
Storytelling System,” IEEE Trans. Comput. Intell. Blood Chapter II: Dwarf Fortress. (2006). Bay 12
AI Games, vol. 6, no. 2, pp. 113–130, Jun. 2014, Games.
doi: 10.1109/TCIAIG.2013.2287297. [26] T. Sylvester, RimWorld. (Oct. 2018). Ludeon
Studios.
[27] Caves of Qud. (Jul. 15, 2015). Freehold Games, Code Fragment 2
South Bend, IN and Berkeley, CA. NPC start and end conditions
[28] “MIARMY | Home.” Accessed: Aug. 29, 2024. Unsatisfied = Definition("Unsatisfied",
[Online]. Available: person).Is(NeedByType[person, __, need],
https://www.miarmy.com/#/ NeedValue[need] == Zero);
[29] “Massive Software.” Accessed: Aug. 29, 2024. People.StartWhen(People.Population[count],
[Online]. Available: count < 5000, RandomPerson)
https://www.massivesoftware.com/ .EndWhen(People[person],
[30] “Defuzz Process - Feishu Docs.” Accessed: Aug. Unsatisfied[person]);
29, 2024. [Online]. Available:
https://basefount.feishu.cn/wiki/GhgswVsrfidf Code Fragment 3
OVkmFcgc1XoYnKA Need assignments
[31] W. Shao and D. Terzopoulos, “Autonomous foreach (var needOfType in NeedTypeList)
pedestrians,” Graph. Models, vol. 69, no. 5–6, pp. People.StartCauses(Add(NeedByType[person,
246–274, Sep. 2007, doi: needOfType, need]).If(
10.1016/j.gmod.2007.09.001. NewNeed[RandomFloat[0.002f, 0.007f],
[32] Q. Yu and D. Terzopoulos, “A Decision Network need]));
Framework for the Behavioral Animation of
Virtual Humans”. Code Fragment 4
[33] W. Wright, The Sims. (2000). Locations and their Advertisements
[34] R. Zubek, “Needs-based AI,” in Game
FairgroundLocations = Parse(
Programming Gems 8, A. Lake., Cengage Learning,
"FairgroundLocations", location.Key,
Florence, KY, 2010.
locationType.Indexed, coords.Indexed,
[35] “Concurrent Interactions in The Sims 4.” ParseFairgroundLocations());
Accessed: Aug. 30, 2024. [Online]. Available: LocationAdvertisements = Parse(
https://gdcvault.com/play/1020190/Concurre "LocationAdvertisements",
nt-Interactions-in-The-Sims location.Indexed, actionType.Indexed,
[36] “Modeling Individual Personalities in The Sims delta, ParseLocationAdvertisements());
3.” Accessed: Aug. 29, 2024. [Online]. Available:
https://www.gdcvault.com/play/1012450/Mo Code Fragment 5
deling-Individual-Personalities-in-The Action satisfies need table
[37] I. Horswill and S. Hill, “Fast, Declarative,
ActionSatisfiesNeed = Parse("ActionSatisfi…",
Character Simulation Using Bottom-Up Logic
actionType, needType,
Programming,” presented at the AIIDE
DefaultActionSatisfiesNeed.Select(
Workshop on Experimental Artificial
pair => (pair.Key, pair.Value)));
Intelligence in Games, University of Utah, Utah,
USA, Oct. 2023.
[38] I. Horswill and S. Hill, “Fast, Declarative, Code Fragment 6
Character Simulation Using Bottom-Up Logic Person action state table
Programming,” in AIIDE 24, PersonActionAt = Predicate("PersonActionAt",
person.Key, actionType.Indexed,
location.Indexed, moving.Indexed);
A. Code Appendix PersonActionAt.Overwrite = true;
PersonMovingTo = Definition("PersonMovingTo",
Other than the two code fragments inside the body of person, location).Is(PersonActionAt[
person, __, location, true]);
this document, the remain fragments in this appendix
contain the entirety of the Simulation code written in
Code Fragment 7
TED/Simulog.
Destinations and Arrivals
Code Fragment 1 Destinations = Predicate("Destinations",
NPCs (people) and their Needs person, coords)
.If(People[person], PersonMovingTo[person,
People = Exists("People", person);
location], FairgroundLocations);
NeedByType = Predicate("NeedByType",
ArrivedAtDestination = Predicate("…", person)
person.JointKey, needType.JointKey, need);
.If(People[person], PersonMovingTo[person,
__], HasPersonArrived[person]);
Code Fragment 8
Arrival state update
PersonActionAt.Set(person, moving, false)
.If(ArrivedAtDestination);
Code Fragment 9
Action timer
ActionTimer = Timer("Action", person);
ActionTimer.StartWhen(ArrivedAtDestination,
PersonActionAt,
ActionTypeInteractionTime[actionType,
count]);
Code Fragment 10
Ready to select action
ReadyToSelectAction = Event("…", person)
.OccursWhen(People[person],
!People.Start[person],
!PersonMovingTo[person, __],
ActionTimer.NotOnTimer);
Code Fragment 11
Action assignment
PersonActionAt.Add[person, actionType,
location, true].If(PersonActionBestScore);
Code Fragment 12
Action completion and reward
CompletedAction = Definition("…", person,
needType, delta).Is(
ActionTimer.TimerFinished,
PersonActionAt,
ActionSatisfiesNeed[actionType,
needType],
LocationAdvertisements);
NeedByType.Set((person, needType), need).If(
CompletedAction[person, needType, delta],
NeedByType[person, needType, needCol],
UpdateNeed[needCol, delta, need]);