drama manager overseeing semi-autonomous agents
TRANSCRIPT
-
8/4/2019 Drama Manager Overseeing Semi-Autonomous Agents
1/12
Interactive Storytelling Essey
DRAMA MANAGER OVERSEEING
SEMI-AUTONOMOUS AGENTS
Interactive Storytelling Essey, topics: drama manager, autonomous agents.
University of Turku
Department of Information Technology
Information and Computation Sciences
Interactive Storytelling
Tommi Laine 84367
28.10.2010
-
8/4/2019 Drama Manager Overseeing Semi-Autonomous Agents
2/12
Table of Contents
1 Introduction ........................................................... 1
2 Stories and characters ............................................ 12.1 Story ........................................................................................ 2
2.2 Non-player Characters .............................................................. 3
3 Restricting the drama management ....................... 4
3.1 Drama manager ....................................................................... 4
3.2 Character behaviour ................................................................. 6
3.3 Character actions ..................................................................... 73.4 Failing believably ..................................................................... 8
4 Conclusions ............................................................ 9
5 References ........................................................... 10
-
8/4/2019 Drama Manager Overseeing Semi-Autonomous Agents
3/12
1 Introduction
Storytelling is a major component in modern computer entertainment in one way
or another. Interactive storytelling enables the user to make decisions that direcly
affect the direction or outcome of the story. Approaches to creating a story that
develops according to the choices of a user can be divided roughly into two
categories: author-centric and character-centric.
In the author-centric approach the plot is woven by providing a drama manager
with a non-branching or a predefined set of braching storylines. The drama
manager is an intelligent, mostly all-knowing agent that manages the virtual world
in order to bring forth a narrative storyline which has been provided beforehand.
The drama manager bends the world according to its predefined will and thus
lowers the believability of the characters. A character might change his personal
goals according to the drama manager's instructions in a way which can be
perceived as schizophrenic behaviour. For example the drama manager might
forbid the player from assassinating an important character in a role-playing game
because that particular character is needed further in the plot.
In the character-centric approach, the characters are autonomous agents, and the
plot rises from interaction between the characters, thus improving believabilityand immersion. All characters act and interact with each other only according to
their personal goals and preferences, without involvement of any god-like level. But
unrestricted character-centric approach also means losing plot coherence, thus
making the content, quality or even the whole genre of the story uncertain.
In this paper I will discuss a way assembled from [1], [2] and [3] to merge the best
characteristics from all approaches by the use of a drama manager and semi-
autonomous agents, while trying to minimize the disadvantages of these
approaches. The goal is to maintain control over the plot while allowing thecharacters to act as autonomously as possible, thus increasing character
believability.
2 Stories and characters
Before character believability and controlling the plot can be discussed in more, the
definitions of story and character need to be determined.
1 / 10
-
8/4/2019 Drama Manager Overseeing Semi-Autonomous Agents
4/12
2.1 Story
All stories can be divided into three levels and I won't be using the terms in the
narrative theory because I find them misleading. I will use the terminology used in
[1].
The first level is the plotwhich contains events that occur in the world, that is, the
content of the story. Not any set of events is considered a plot. A plot must be
consistent so that the sequence of events is natural and in accordance with the
story world. The events should also have an intriguing structure so that the plot is
formed into the form of a conventional story, i.e. a beginning, a climax and an
ending.
The second level is a narrative that is an abstract representation of the plot from a
particular point of view, e.g. from the eyes of a playable character. Narration in a
chronological order of events is called linear narrative whereas the sequence of
events told in non-chronological order is nonlinear narrative.
The third level of story is the presentation. Presentation determines the medium
through which the narration is realized e.g. text, speech or animation.
Figure 1 contains example with a one thrilling story about a lumberjack, two bears,
revenge and misunderstanding. The example shows that the narrative andpresentation levels are similar, especially in character-centric action where
characters' actions make the plot, but the presentation can be done differently
from the narration like in the lowest cutscene element in the presentation pile.
The example also shows situations where the narrative does not have an element
of the first confrontation with the bear but presentation includes a immersive
sneakpeek to the past. This is done with a cutscene while another avatar tells the
lumberjack avatar what happened. In both narrative and presentation levels the
lumberjack has the same knowledge, but presentation level implements anelement of non-linear storytelling.
2 / 10
-
8/4/2019 Drama Manager Overseeing Semi-Autonomous Agents
5/12
2.2 Non-player Characters
Simulation of social interaction can be modeled through a collection of non-player
characters (NPCs) that appear intelligent, motivated and reactive [2]. Usually the
underlying goal storytelling systems try to achieve is not to have full artificial
intelligence reasoning agents, but agents that appear to be intelligent, motivated,
emotional and capable of learning.
This approach is called "broad but shallow" approach e.g. character might have a
goal to go and eat food but eating food would not be a requirement for the
character to stay alive. This approach will also be used in this paper. If left alone
without the direct control from the drama manager, characters should be able to
act lifelike without external instructions, pursuing their own goals to maintain high
level of character believability.
3 / 10
Figure 1: Example of the story levels.
Lumberjack fighting
with the bear
Lumberjack knocked
unconscious
Lumberjack wakes
up with amnesia
Lumberjack decides
for revenge
Lumberjack fights
again with wrong bear
Lumberjack kills
the wrong bear
A bear sees
a lumberjack
Lumberjack is told
what has happened
Lumberjack wakes
up with amnesia
Lumberjack learns
what has happened
Lumberjack decides
for revenge
Lumberjack fights
with a bear
3D world
renders lumberjack
avatar waking up
Lumberjack avatar speaks
with another avatar
Lumberjack avatar speaks
speaks with himself
Lumberjack kills
the bear
Cutscene from the
eyes of the dying bear
Lumberjack avatar fights
with a bear avatar
Plot Narrative Presentation
Cutscene of the past events
told by another avatar
-
8/4/2019 Drama Manager Overseeing Semi-Autonomous Agents
6/12
But, in this method of overseeing, the drama manager is still needed to ensure that
a non-player character does not perform actions that might totally ruin the plot,
that will be discussed further in the essey.
3 Restricting the drama management
To achieve both a coherent plot and believable characters, a drama manager can be
controlling the behaviour of semi-autonomous agents at some degree, making
them follow the plot.
3.1 Drama manager
The drama manager can be set to three different level of detail [1]. Firstly you can
predefine all branches of actions the characters can choose from to the drama
manager. Secondly you can have the drama manager to ensure the occurence of
some essential events. Or thirdly you can tell the drama manager to make sure
small narrative fragments are regularly inserted.
As we are using a drama manager adopted from the author-centric approach, the
main rising concern is the belieavability of semi-autonomous agents that must be
dealt with.
There are total of three ways the drama manager should be allowed control the
plot in this method [1] summarized in figure 2. To begin with, the drama manager
can subtly introduce a new enviromental element like a new character or object
into the story world. Subtly meaning that items do not appear out of thin air. We
will call this enviromental control.
Secondly the drama manager can use motivational control, used in example of the
next paragraph i.e. enforce a new goal to a particular character.
And thirdly the drama manager can use proscriptive control, disallowing an action.
When a character wants to initiate an action, it will always ask the drama manager
is it a legal action. I should also be noted that the drama manager has no
prescriptive control, meaning it can not force a character to act instantly, just
change the underlying goals of the characters so the believability does not suffer
from the overseeing.
4 / 10
-
8/4/2019 Drama Manager Overseeing Semi-Autonomous Agents
7/12
To support proscriptive control of the drama manager, there should also be createdcausal links between the events [2]. The causal links are used to spot a sequence of
events made by the interacting player that may threaten the plot coherence. e.g. if
an assassin has a mission to kill another character by stabbing, the assassin needs
to have a bladed weapon. In this kind of situation, the causal links come into the
play. The inconsistency of the assassin not having a bladed weapon can be circled
around by generating new event where the bladed weapon is not needed.
For each point of inconsistency that will arise, firstly the drama manager will try to
repair the causal link in question by its three ways of control defined above e.g.giving the assassin a knife in a natural way.
If the repair by fullfilling the event requirements is not possible, the drama
manager attempts to remove any events that were dependent on the threatened
causal link and then repair the plot by filling in events the drama manager feels
required to restore the plot coherence e.g. assassination by strangling and this
change in the events is taken into account in all future events in the plot.
5 / 10
Figure 2: Three ways a drama manager of this method can control the flow
of the events.
Character
Character
Character
Drama manager
'I am feeling
hungry...
'Can I eat this?' 'Yes.'
Enviromental
control
Motivational
control
User
User's point of view
Proscriptive
control
-
8/4/2019 Drama Manager Overseeing Semi-Autonomous Agents
8/12
Lastly, if all else fails, the drama manager selects new goals and rebuilds the plot
e.g. the assassination never happens and the outcome is different.
All in all, if the user performs an action that causes an inconsistency that threatens
the plot coherence, the drama manager looks up the appropriate branch in the tree
of events and seamlessly begins directing the semi-autonomous agents based on
the new plot. See figure 3 for example where the red arrows mark the plot
reconstructions that the drama manager does and black arrows are causal links. It
should also be noted that the inconsistency of casual links can only happen fromthe actions of the interacting user, because NPCs always ask the drama manager
before they act.
3.2 Character behaviour
I propose using the method used in [2] and [3] for the model of character
behaviour while introducing a level of priotiy to the NDBs. Character motivations
6 / 10
Figure 3: Plot reconstructions and causal links.
has(bladed weapon)
can not obtain(bladed weapon)
Initial state
Stab(target)
does not have(bladed weapon)
Strangle(target)
Acquire(knife)
Intermediate stateOutcome(death by stabbing)
Outcome(death by strangling)
Intermediate state
Outcome(nobody dies)
knows(martial arts)
has(bladed weapon)
can not know
(martial arts)
-
8/4/2019 Drama Manager Overseeing Semi-Autonomous Agents
9/12
are divided into two kinds of behaviour: local autonomous behaviours and
narrative directive behaviours.
Local autonomous behaviours (LAB) are sequences of actions that make the
character seem humanlike. e.g. eating lunch, chatting with people, working. The
idea is same as of objects in object-oriented programming. The character itself only
knows the contents of his LABs and others don't have to know the implementation
of the actions. Even if the LAB contains interaction with another agent, the target of
interaction must only provide an applicable interface for the interaction. Character
will continue doing these, at some degree repetitive and generic, tasks to create
illusion of lifelikeness.
Narrative directive behaviours (NDB) are sequences of actions that advance the
plot. They are behaviours enforced to the characters by the drama manager and are
initiated when the change of underlying goals does not cause a stir in the
believability.
For example if a character is eating (a LAB), the playable character is present and
in the plot there is an event for this particular character to commit a assassination
(a NDB), the character would not suddenly stop eating and run to the murder
scene. The character would either finish eating or, if the priority of NDB is high,
provide an excuse or request an excuse from the drama manager to stop eating e.g.
make a personal remark: "Oh! I'm so full already!" or receive an incoming call
generated by the drama manager, allowing him an excuse to leave the scene.
On the other hand, if the priority of the NDB would be low, the behaviour would
only be initiated when the character is figuring out a new goal because the
previous goal was achieved.
Both LABs and NDBs need to have requirements i.e. resources they require.
Requirements can be location, object, emotional state, player not present and so
on. When the requirements are met and if the drama manager is giving a green
light according to postscriptive control, the action sequence of the behavior can be
initiated.
3.3 Character actions
Non-player characters have their own set of actions they can execute. Actions can
be for instance picking up an object, using an object to another object, eating an
7 / 10
-
8/4/2019 Drama Manager Overseeing Semi-Autonomous Agents
10/12
object i.e. destroying an object with animation and so on. When a character has
behaviour provided by motivational control of the drama manager (NDB) or a
behaviour in character's of repertoire of generic believability supporting
behaviours (LAB) to pursue some goal, the character wil try to accomplish the goalwith actions it can do.
For example there is a character that has a LAB with requirements of player being
present, the character not having eat in a while and the character having food like
banana. If these conditions are met the character will ask the drama manager if it is
legal to eat the banana. If the action is confirmed by proscriptive control of the
drama manager, character can eat the banana accompanied by a munching sound
and an animation.
3.4 Failing believably
This all is making sense on a paper but because we are discussing interactive
stories, the drama manager does not always know every possible thing yet what
the characters should be forbid to do. Proscriptive limits may become activated
only after some particular event in the plot. Next we will discuss how to handle
situations where a character has adopted a behaviour and initiated some actions
towards a now illegal goal.
I will be adopting a concept of failing believably from [3] to tackle this problem. In
this kind of situation, we have a character with a sequence of now illegal actions
but has not yet reached the goal e.g. an assassin character has located himself to
the murder scene but now the drama manager's current plot structure requires the
target of the assassin to stay alive a while longer. As the assassin ask the drama
manager if it is legal to jump from the shadows and strike the dagger to the heart of
the target, the drama manager gives the assassin a behaviour to fail the
assassination and flee the scene.
Of course there would not be any problem in the first case if the user experiencing
the story does not know anything about the assassination, but for the same of the
example lets determine that the user is a spectator in some way. Even if the user
knows about the assassin leaving for a mission to assasinate another character,
there is need to generate a backup story why the assassintion did not occur.
If this kind of method of automated failing would be implemented in an interactive
8 / 10
-
8/4/2019 Drama Manager Overseeing Semi-Autonomous Agents
11/12
storytelling system, it could have alternative uses too. For example the failing
believably can be used as a tool to boost the believability of a particular character
as humans make errors. Althrough even if does not differ from a prescripted
sequence of actions much, a template for the believable failing could turn outfruitful.
4 Conclusions
This paper describes a way to allow a drama manager to control agents of an
interactive story so that the character believability stays still reasonable. The
method is assembled by merging the works [1], [2] and [3] with some own twists.
The agents won't be fully autonomous and the drama manager won't have absolutecontrol of the events in this approach but there is alot more flexibility than the
extremes of author- and character-centric approaches. The agents can not be fully
autonomous if there is a desire to have some control over the plot and the drama
manager can not have absolute control if you want to maintain majority of the
character's believability. If the characters are allowed to do anything they desire,
they might break links in between the events of the predefined plot so we allow
some control tools for the drama manager.
9 / 10
-
8/4/2019 Drama Manager Overseeing Semi-Autonomous Agents
12/12
5 References
[1] Theune, M.; Faas, S.; Nijholt, A.; Heylen, D.; (2003)
The Virtual Storyteller: Story Creation by Intelligent Agents.
Proc. of the 1stInt. Conf. for Interactive Digital Storytelling and Entertainment.
[2] Riedl, M.O.; Stern, A.; (2006)
Believable Agents and Intelligent Story Adaptation for Interactive
Storytelling,
Proc. of the 3rdInt. Conf. For on Technologies for Interactive Digital
Storytelling and Entertainment
[3] Riedl, M.O.; Stern A.; (2006)
Failing Believably: Toward Drama Management with Autonomous Actors in
Interactive Narratives
Proc. of the 3rdInt. Conf. For on Technologies for Interactive Digital
Storytelling and Entertainment
10 / 10