User:Sz373 cs/DRAFT

Introduction
 Human-Aware Architectural Design  is an one concentration of 3D modeling in Computer Graphics. The term is used when we design architectural models like a house or a city which can detects its context - human. Sometimes it utilizes artificial intelligence algorithms. Human has many sensors to detect out side world,at the same time, in computer graphics area, we also need some methods to help agents detect other objects and find the way to destination faster. On the other hand, a better design for placement of terrain or obstacle will lead to a better crowd agent evacuation. The layout of a building, real or virtual, affects the flow patterns of its intended users. So the human-Aware architectural and urban design is of great importance to the architecture, robotics, urban simulation, game development and other communities to explore the configuration space of environmental elements for a variety of reasons and applications.

Crowd Evaluation
Use several methods such as density measures and fundamental diagram based comparisons to simulate the crowd evaluation and compared with real world situation to improve performance.

Optimizing Crowd Simulation Parameters
Selection of a steering algorithm's parameters can dramatically influence the performance and behavioral patterns of the aggregate crowd dynamics. There proposes solutions for automatically fitting a steering algorithm's parameters to minimize collisions, minimize evacuation times, or match recorded data.

Game Level Optimization
Use several methods such as evolutionary approaches and combinations of objectives to find interesting variations in the game level, give game maximize "fun", formulates a parameterization of the game level and evaluates each game level’s expected difficulty.

Arichetectural simulition
UrbanSim is an open source urban simulation system designed by Paul Waddell (University of California, Berkeley) and developed with numerous collaborators to support metropolitan land use, transportation, and environmental planning. It has been distributed on the web since 1998, with regular revisions and updates, from www.urbansim.org. Synthicity Inc coordinates the development of UrbanSim and provides professional services to support its application.

Architectural optimization
Systematically observe the sensitivity of steering algorithms and its parameters, and the parameters of environment elements on the optimization results as well as evaluating additional environments.

Steering Algorithms
There are three established steering algorithms that represent a range of different steering approaches.

a. ORCA: an efficient and widely used technique that uses reciprocal velocity obstacles for collision avoidance.

b. PPR: a hybrid approach that uses rules to combine reactions, predictions.

c. SF: a variant of the social forces method for crowd simulation.

Main modules in human-aware architectural design
Main modules you will need to create a human-aware architectural design: - User and user group: they provide all user information which can be transferred to avatars or agents in the system. This is one of the major missing parts in conventional architectural design. In order to establish the model capable of dealing with human-space interaction or human-centered design, this module is crucially needed. This module also includes behavior information empowering an ability to let agents or avatars perform predefined actions in certain contexts. - Location: these entities provide the both semantic and geometric relationship among objects, spaces and users in a place stored as a location. - Activity: the concept of activity embodies the characteristic of ‘place’ since it contains and combines user, place, location and action. Like space in building data model, activity is essential for connecting with other data models. - Spatial domain and type: this inclueds the information of each type of building and space having different attributes and constraints. Serve as libraries during the design process; it helps architects to create places rigorously.

AMR
AMR: Adaptive Mesh Refinement (AMR) adaptively samples the discretized optimization search space, a constrained region in the environment. AMR is a technique commonly used in the simulation of turbulent hydrodynamics.

AMR affords tunable optimization results for crowd aware design of environments by adaptively discretizing the optimization space. This facilitates an interactive authoring-optimization loop where designers are given the results of optimized environments which they can use as iterative feedback during the design process.

Each environment element is optimized individually given an initial best guess for all elements. After each element is placed, the crowd flow data is updated and the mesh is refined. Based on results from previous literature, our granularity heuristic is crowd density which focuses sampling areas of heavy crowd flow.

ISAB
Iterative Selection AMR with Backtracking (ISAB) iterates through all parametrized environmental elements Ep, placing one on each iteration. At each iteration the environment is simulated and a density histogram is produced. A default flow value is computed by executing a simulation without the current element. This histogram is then used to adapt a space-partitioning structure, or mesh. The nodes of this mesh are used as sample points at which to reinsert and place the current element. The resulting scenario is then simulated, and the flow value at the current node is stored. The index with the maximum value is then the optimal selection on this iteration. A back-tracking failsafe ensures completeness in that we avoid being stuck in a local minima after an initial bad selection. The amount of backtracking for any given element is also tunable (2 iterations here). The condition for backtracking is a reduction in flow in the current iteration and some small probability. It should be noted, however, that in our experiments the backtracking condition was never triggered. We expect that in larger more complicated experiments the possibility to backtrack will avoid local minima. Our algorithm then is greedy iterative selection with the ability to back out of bad solutions.

Application
Human Aware Architectural could be used in many areas such as robotics, crowd evaluation and computer & video games.

Example
The obvious example that can be found in situation is simulation games such as ‘The Sims2’ in which each user performs ordinary tasks imitating the life in real world. The game playing depends on emotional and behavioral characteristics of multiple users through complex scenarios. - task-based design, scenario-based design and performance-based design - therefore, can explain why the level of presence in a situation simulation game is high enough to enable game players to immerge and to enjoy the interaction in virtual environment. Apart from these studies, a number of outstanding VR simulation platforms have been developed revealing the same tendency.

Introduction
Computational models of narrative describes the ability for computers to tell, understand and respond affectively to stories. It is an interdisciplinary field, combining the human experiences of narrative and storytelling with the computer science subfields of artificial intelligence, machine learning, human-computer interaction and computer animation. The field seeks to instill narrative intelligence in computers, enabling them to communicate, craft and understand human stories. The field is nascent, with much of the research done theoretical, but presents a variety of unique machine learning and computer animation challenges.

Narrative intelligence has long been understood as one of the abilities that sets humans apart from other animals. . Research in computational narrative intelligence has sought to create computational intelligences that can answer questions about stories, generate fictional stories and news articles, respond affectively to stories, and represent the knowledge contained in natural language narratives. The obvious application of fictional story generation is entertainment. On-demand narrative generation can maintain a continuous flow of novel content for users to engage with while customizing the content to individual preferences and demands. Finally, narrative can be used to explain the behavior of artificial intelligences. Any process or procedure can be told as a narrative, so it follows that an AI can describe the means by which it came to a conclusion or the reasons why it performed an action by couching its explanation in narrative terms. As part of a explanatory process, narratives can convey counterfactuals—what would have happened if circumstances had been different.

Background & History
Storytelling is an important part in the life of humans to communicate, entertain and teach. Narrative is a natural and intuitive way of describing the semantics and status of a situation. It engages humans' cognitive abilities, allowing historical and causal knowledge to be interpreted and passed down (Finlayson, et al). We can make our narrative by our experience, but computers can not create a story without humans. With the development of video games and artificial intelligence, narrative will be more and more important in computational models. Nowadays, most games provide at least some stories of pretense to guide players. Thus, computing narrative is important to have machines react effectively to the situation or task at hand. Creating more lifelike NPC's in digital worlds creates greater immersive experience for the player. Besides video games, digital story telling has also been found to treat mental disorders, simulate research environments (crowd analysis, population control, etc.) and effectively teach. Gameplay with meaningful story can help someone deal with their emotional issues, such as stress, depression, etc. It becomes an outlet for them and can persuade the player to adopt a different perspective on their situation. Simulations in research require agents to interact with one another and understand series of events so they can accurate respond. Teaching in simulated environments could prove to be more effective due to the ability to create scenarios for the player to interact with.

Propp's Morphology of a Folktale was an effort by a 20th century literary scholar to create a hierarchy and structure of the standard folktale, as a basis for analysis. (Propp, 1968)

Schank modeled cognition in terms of scripts, mental templates that the mind executes based on the situation and environment. (Schank, 1977)

Dallenbach, another literary critic, attempted to describe a specific narrative pattern, that of a play-within-a-play. (Dallenbach, 1989)

Pearson focuses on the protagonist of stories, defining a series of archetypes for categorizing goals and motivations. (Pearson, 1989)

Jerome Brumer described the characteristics of narrative that separate it from random phrases in a language (Bruner, 1991)

Kerstin Dautenhahn, of the University of Hertfordshire, proposed the Narrative Intelligence Hypothesis, which views narrative as a necessity in the development of complex social structures. (Dautenhahn, 2001)

Mateas and Sengers coined the term narrative intelligence (NI) to describe the cognitive ability of capturing knowledge through narrative. (Mateas, 2003)

Cheong generates stories judged to be suspenseful by modeling the reader’s reasoning about limitations and conflicts involving a protagonist’s goals. (Cheong, 2007)

1.Narrative Structures
Traditional narrative focuses on three concepts: the story, the storyteller, and the audience. A story is an account of incidents or events. The audience is the body of listeners or spectators that experience a story. The storyteller acts as the interface between audience and story.

In computers' role, the software offers the writer feedback regarding where its representation of correctly spelled words or proper grammar do not match up with the writer's story. These actions compose a feedback loop between writer, the story, and the computer. The story is described to the computer in a way which allows the computer to understand it enough to facilitate simple manipulation. There are two alternative approach. First is A Knowledge Based Approach, it is well known in the AI community that one of the weaknesses of knowledge-based AI is that its structures become brittle when faced with a dynamic problem domain or any problem domain which it was not specifically designed to handle. Second is A Behavior-Based Approach, this approach is described well by Pattie Maes in her paper entitled Behavior-Based Artificial Intelligence, where she compares and contrasts the two forms. In it she lists characteristics which typify the knowledge-based and behavior-based approaches.

In agents' role, this approach taken with Agent Stories is to assemble narratives in either textual or QuickTime movie form by making use of the three components of computational storytelling: the structure of the narrative, the collection and organization of story pieces with some representation of their meaning, the navigational strategy through that collection of story pieces, with style and purpose.

2.Challenges
Usually there several challenges related to computational narrative.

First, computers should understand some elements that are assumed to be commonly shared knowledge among humans.

Second, it is necessary for both narrative understanding and narrative generation.

Third, decoding the meaning of metaphors and metonymy requires high-level semantic comprehension of the narratives.

Forth, there will be a large training set for computers to create stories.

Fifth, there are multiple environmental agents that interact with one another to create meaningful narrative.

Sixth, reasoning with computerized agents to persuade certain development. How to persuade effectively with a story? How to make the agent determine its own judgement given information?

==== 3.Interactive Narrative ==== The motivation behind the paper, "Requirements for Computational Models of Interactive Narrative" by Nicolas Szila, is to understand the requirements for an interactive narrative. This means what rules we must follow to differentiate it from a regular narrative. Then it explores how to achieve those requirements and find the differences between existing models for story generation.

This is an important topic because true interactivity is scarce. "Interactive storytelling" in video games is limited. The games' responses to the user's actions are limited by what is preprogrammed. It is not "true" interactivity. The computational models would need to be compelx to accommodate player's choices into the story. Every action by the player needs to affect the world long term so the effects aren't only superficial.

The definition of interactive narrative is an entity that contains an initially developed core story which is the center of the world's drama. It shows social narrative: characters are easy to emphathize with and engage the audience to want to explore further. The narrative is deeply integrated with the choices that the user makes. There will be plot and character development outside of the main story. Overall, the story is dynamic, changing depending on the user's actions.

There are eight requirements that must be satisfied to meet the criteria of a interactive narrative. The entity has genericity: many possible actions by the world environment; causality: the player's actions cause actions to be taken; characters: main characters should be complex and promote empathetic viewing; transformation: it must be possible to go from an initial state of unreached goal to final state where the goal is reached; unity of action: all actions are organized around a single line of action (possible actions stem from initial action); narrative sequence: actions in a narrative are spread out in a sequence; message: narrative contains an intenton or message to convey to the audience; emotional involvment: capable of triggering the audience's emotions to create enjoyment and story understanding.

Interactive narrative and story generation are often confused together but differ on several points. Having the user capable of acting upon the environment creates development of feelings for the world. Passive viewing of a narrative creates tension from lack of being able to intervene. Role-playing games can give a player a strong narrative experience but it would give an outside viewer a boring point of view. Sometimes, it's only entertaining when you are part of the experience. It would be worth looking into to move from narratology to psychology and consider the process through which the user engages with the narrative.

There are existing forms of interactive narrative that we couuld base our computational models off of: role-playing games, improvisational theater, and oral storytelling. The problems with this, however, is that there are limited theoretical investigations and that they are difficult and complex to detail. Furthurmore, how would we go about adapting a model of human to human activity to human to artifact (system).

To achieve a model, it is necessary to explore interactive positive cases: cases when narrative events seem valuable in an interactive narrative. We should also explore linear negative cases: cases when narrative events that are valuable according to "linear narrative" laws seem not relevant in an interactive situation. Most of all, the player has to believe that the computer has human-like intelligence.

In conclusion, the story lived by the user as an active participant is fundamentally different from a viewing audience. A broad model of narrative is needed to be flexible in order to adapt to the user's choices and random probabilities.

1.Artificial Intelligence in Computer Games
Game AI refers to algorithmic techniques to augment the player's experience in computer and video games. The goal of Game AI as a discipline is to produce the illusion of intelligence in the behavior of Non-Player Characters (NPCs—the opponents, companions, and other entities in the virtual game world) in the virtual world of the computer game. Almost all modern computer games utilize some form of artificial intelligence, making games the largest class of commercial product through with public regularly comes into contact with artificial intelligence. Laird and van Lent (2001) put forth an argument for AI in computer games as an academic pursuit.
 * "Human-Level" AI
 * Better games
 * Supporting Game Development Practices
 * New Experiences

2. Requirements for Computational Models of Interactive Narrative

 * Genericity: To ensure the large space of possible action mentioned above, going beyond a large space of combination of predefined actions, actions must be defined in a generic manner, with variable that can be instantiated among several specific objects in the computational fictional world. Typically, a predicate-based formalism enables this level of genericity. Higher order predicates increases the level of genericity.
 * Causality: These actions must be arranged in a causal manner, as causality is one of the main characteristics of narrative. Defining the type of causality that exists in narrative should be far from straightforward. Causality of characters' own plans, causality of actions between characters and between character and story events, causality of the discourse are all valid forms of causality that a computational model of narrative for Interactive Narrative will implement. It is certainly the case that multiple forms of causality intervene in a single narrative, therefore, it is necessary to investigate further what is narrative, which cannot be reduced to a mere causal chain of events.
 * Characters: These actions involve characters. This obvious feature should be reminded because characters are not only the subject of actions, but also the focus of the audience's attention and interest. Therefore, a full model of narrative should ensure that main characters be perceived as rich and complex entities and be able to promote empathetic viewing.
 * Transformation: A narrative must concern a transformation. To achieve a goal, for example, is a transformation that goes through from an initial state in which the goal is not reached to the final state where the goal is reached.
 * Unity of action: In a traditional narrative, all actions are organized around a single line of action, which is called the unity of action. In modern narrative, this principle is accommodated to take into account episodic forms and subplots. However, for the sake of story understanding and cohesion, the number of lines of action should be limited and controlled.
 * Narrative Sequence: Main actions in a narrative are usually spread in a sequence that consists of the initial information on the possibility to perform an action, the influence regarding the performance, the performance itself, and the final sanction.
 * Message: Going beyond the story level, narrative should also be taken into consideration from its pragmatics point of view, that is, at the discourse level. Each narrative contains a general intention/message that it aims to convey to its audience. It is typically contained within the handling of a system of values according to which story events are explicitly or implicitly evaluated.
 * Emotional involvement: This is also a key dimension to make a narrative successful. Triggering audience's emotions are not only an enjoyable feature of the narrative, but also a necessary condition for story understanding.

3.Interactive Narrative use AI
Interactive Narrative is a form of digital entertainment in which users create or influence a dramatic storyline through actions, either by assuming the role of a character in a fictional virtual world or by issuing commands to an autonomous NPC. In computer game, story line have this properties: However, storylines in most commercial computer games cannot be created or significantly influenced by the player. If the storyline in a commercial game can be influenced by the player it is because a small number of choice points—called story branches—have been hard-coded into the game.
 * Provides a contextual against which the actions, tasks, and characters make sense.
 * Motivates the player to perform actions
 * Creates transitions between various task and activities

AI approaches to interactive narrative generally fall into one of two classes. AI need to balance the authorial intent and player intent in the context of the story telling which is a challenge for interactive narrative.
 * Emergent narrative systems
 * Drama management systems

4.Plan Recognition and Synthesis
Computational narratology explores narratology – the art and science of storytelling – from the point of view of computation and information processing. One of the key features of understanding narratives requires understanding characters and characterizations. Traditionally, we take an intentional agent based approach to understanding characters. We take each character as having a goal, and take the computer's job to be to figure out and then execute that goal. There have been several heuristics to do this. One of the earliest ones was to apply what's called a Plan Applier Mechanism, which essentially interpreted every action in the story as instantiating the goal of a character, and the computer would draw upon a "knowledge pool" to determine what each action of the character meant. Thus, in an ideal world, if the computer saw "John went to the bank," the computer would interpret the goal as "John needed money." While the heuristic was sound, implementation is computationally incredibly difficult, as it requires the computer to draw from an infinitely large knowledge pool as well as make non-deductive reasoning observations.

This heuristic evolved several years later, with another heuristic that essentially only needed a character's initial state and goal state as inputs, and would draw upon a library of prebuilt scripts to achieve that goal. While the heuristic has changed in implementation, the idea has remained constant. Current problems in goal synthesis involve replanning and outcome evaluation. Replanning essentially involves dynamically reacting to a change in the narrative, and outcome evaluation is used in machine learning techniques to generate new narratives. In video games for instance, if a player chooses one outcome, then outcome evaluation would evaluate what the character did to determine the next phase in the story.

One of the most sophisticated planners used today is called the IPCOL planner: the intent-based partial causal order link planner. This uses partial-order planning to leave outcomes as open as possible. How this works is the system follows the "Principle of Least Commitment," where the next phase is the one that leaves open the most outcomes. IPCOL makes this ordering based on intent; thus, it constructs an optimization problem that tries to optimize "intent" (i.e. what the character wants to do) with a plan that leaves open as many possibilities.

5.Open problems for Interactive Narrative
Besides there are also some open problems for interactive narrative, and these problems are related to how to make computational systems reason about narrative and manage players' interactive in virtual world.


 * Story knowledge representation ：How does the drama manager know what a “good” player experience should be? The drama manager is a surrogate for the human author.


 * Real Time Adaptation ：A drama manager must respond to the player’s actions in a way that neither diminishes the player’s perceived agency nor violates authorial intent (Magerko 2005). Drama managers must search for the best future story experience for the player based on the player’s actions, preferences, and current story world state.


 * Story generation ：A more general solution to Interactive Narrative is to instill the drama manager with the ability to automatically generate novel branches.


 * Authoring ：AI systems require knowledge. Regardless of whether the drama manager is manipulating plot graphs or constructing novel story branches using a story generator, some amount of knowledge must be authored by the human author.


 * Believable Character Agents : A believable character is a character in an Interactive Narrative that does not act in a way that breaks the player’s suspension of disbelief


 * Player Modeling : A drama manager may also act as a surrogate for the player, by modeling and predicting player preferences for different story experiences.

1.Computational Narrative Intelligence
Narrative intelligence is one of the abilities that sets humans apart from other animals and non-human-like artificial intelligences. The obvious application of fictional story generation is entertainment. On-demand narrative generation can maintain a continuous flow of novel content for users to engage with while customizing the content to individual preferences and demands.Computational narrative intelligences can also create plausible sounding—but fictional—stories that might happen in the real world. Plausible real-world story generation can be used to generate virtually unlimited scenarios for skill mastery in training simulations.Computational narrative intelligence also brings computers one step closer to understanding the human experience and predicting how humans will respond to narrative content.Narrative intelligence offers a useful account of different ways in which the relationship between narrative and computing can be viewed. Originally, the theme of narrative intelligence drew upon the following streams of influence: cognitive science and artificial intelligence, literary theory, art, drama, media studies, narrative psychology and sociology, user interface theory, software engineering, and social computing.Some observations made by narrative intelligence researchers Marc Davis and Michael Travers break down the important differences between the relationship with texts held by literary theorists and computer scientists. In some cases they did not account for literary theorists’ analysis and creation of computational texts so the account of their observations below has been slightly supplemented. Some of these differences, in a form they admit as somewhat caricatured, are:
 * Reading natural language texts: The literary theorist closely analyzes style, implicit meaning, and structure while the computer scientist values quick extraction of core concepts and utility.
 * Reading programming language texts: The literary theorist typically does not address programming language texts (though recent theorists such as Espen Aarseth and Ian Bogost have begun to do so) while computer scientists are able to develop detailed understanding of text functionality so that it can be executed or appropriated for new text production.
 * Writing natural language texts: The literary theorist considers this to be the primary form of text production. Typically this production is done to analyze other texts. Stylistic innovation is rewarded. Computer scientists often produce these texts secondarily to programming language texts. They often are analyses and documentation of programming languages texts. Stylistic innovation is not highly rewarded, while clarity of exposition is valued.
 * Discussing and presenting work: The literary theorist uses speech as a rhetorical art form where innovation, cleverness, and complexity are rewarded. Computer scientists often value lack of ornament and simplicity of exposition. Demonstration of implemented work is highly valued.

Work in NI has drawn on conceptions of narrative from many of these sources. Researcher have explored lots of topics relevant to NI. Applications of narrative intelligence Related Algorithms
 * Art
 * Psychology
 * Cultural studies
 * Literary studies
 * Drama
 * NI is Humanistic AI
 * Narrative Interfaces
 * Narrative Agent Design
 * Agents that Use Narrative Structure
 * Support for Human Storytelling
 * Story Database Systems
 * Story-understanding Systems
 * Storytelling Systems
 * Interactive Fiction and Drama
 * Narrative for Meta-analysis
 * Story Telling: The obvious application of fictional story generation is entertainment. On-deman narrative generation can maintain a continuous flow of novel content for users to engage with while customizing the content to individual preferences and demands. One may imagine serial novels, serial scripts for TV shows and movies, or serial quests and plotlines in computer games.
 * Question answering: Question-answering is a way of verifying that a computer is able to understand what a human is saying. One prerequisite for narrative question-answering is a better understanding of how to represent the knowledge contained in natural language narratives.
 * Plausible sounding: The generation of plausible real world stories provides a strong, objective measure of general computational intelligence. Plausible real-world story generation can be used to generate virtually unlimited scenarios for skill mastery in training simulations. Computational narrative intelligences could engage in forensic investigations by hypothesizing about sequences of events that have not been directly observed. Virtual agents, such as virtual health coaches, can appear more life-like and create rapport with humans.
 * Predictor of human behavior: Computational narrative intelligence also brings computers one step closer to understanding the human experience and predicting how humans will respond to narrative content.
 * Event Learning: Event learning is a process of determining the primitive units of action to be included in the script.
 * Script Learning：This process produces a general model of expected event ordering for the given situation.

2.Machine Enculturation
Machine enculturation is the act of instilling social norms, customs, values, and etiquette into computers so that they can more readily relate to us and avoid harming us or creating social disruptions. However, manually encoding a comprehensive set of values or rewards in order to create social cultural behavior for any sufficiently complex domain, such as the real world, is intractable. So we can use collected works of fiction by different cultures and societies to improve artificial intelligence learner. Machine enculturation may give us a way forward toward achieving artificial intelligences that understand humans better.

3. Adaptation to stimuli
With a general complex adaptive system model, there is accounting for other agents in the system. It allows for inter-agent activity, which greatly increases system reliability to produce accurate simulations. This is because in realistic settings, the environment can affect itself through agents. The feeling of the world adapting around the player creates for more immersive environments. One such example of using this solution is exploring marine ecosystems for the purpose of researching predator-prey populations. For this, it is completely necessary to have a functioning model for developing interactivity between the agents. A predator must hunt prey and the prey must avoid the predator. Only then could accurate results be gathered.

4. Persuasive Stories
Although the solution itself is not clear, there are several suggested ways to pursue. One of which is to use agent-based practical reasoning that accounts for stories. The goal is to create a self-sufficient system that can make its own choices depending on the context given. This is a great tool specifically for teaching purposes. With clear reasoning, characters can persuade one another to understand a certain topic. These issues are also centered in the fields of psychology and sociology.

Subtopics
Digital storytelling

Interactive narrative for use in applications such as video games, where the story and gameplay varies depending on your choices. Many of the previous topics touched upon would be used here, such as persuasion, agent adaptability, and proper reactions.