Intelligent virtual agent

from Wikipedia, the free encyclopedia

Intelligent virtual agents are autonomous, graphically modeled and animated characters in a virtual environment that have artificial intelligence . For a lifelike effect, they interact with their surroundings, with each other or with people using natural modalities such as language, facial expressions and gestures. The perception and action in real time enables them to be part of a dynamic, social environment.

Delimitations

  • Differentiation from animated films: Intelligent virtual agents are rendered in real time due to interactivity . In contrast to real-time rendering , the film industry uses the offline rendering principle to produce animated films. With the offline rendering principle, it is determined beforehand what is to be rendered. An image is then calculated, which can take several hours.
  • Computer animated characters in movies are not intelligent virtual agents because their behavior is predetermined by a script.

Other names

  • Software agent : This is the name of an agent that can independently carry out actions in a virtual environment in order to achieve its goals. The agent's appearance can be shown graphically where there is a strategic advantage.
  • Embodied Agent: An embodied character is a system in which a character is graphically represented in a virtual world. The character can be represented in the form of a human or an animal and helps users accept it.
  • Avatars are controlled by the player and have no autonomous behavior.
  • Non-player characters (NPC): Are characters controlled by the computer according to the rules of artificial intelligence , who are either friendly or neutral towards the player.

Areas of application

"Virtual Sheep Dog"

In addition to computer games for entertainment, intelligent virtual agents are also used in e-learning and for anthropomorphic user interfaces.

Computer games

In computer games, they are used as non-player characters to convey information, tips or tasks to the player about the game, or simply to make the game world seem more lively. In addition, as teammates they can help the player to solve tasks or, as opponents, they can pose challenges to the player. In addition, they can step out of the player-controlled role and independently carry out decisions made by the player.

E-learning

Intelligent virtual agents are used in the e-learning sector as educational agents such as B. used as a tutor , mentor , teacher, learning partner, etc. to provide the user with a wide variety of learning content didactically. They serve as presentation agents on websites and at information terminals to present information to the user.

Anthropomorphic user interfaces

Virtual agents serve as an interface between a system and the user. The user can instruct the agent to perform various tasks. The virtual agent can develop into an operational program.

Applications

  • Project Milo: A virtual child developed by Microsoft that can be interacted with through speech and gestures.
  • Façade: In this software, the player can interactively take part in a conversation between two married couples using keyboard inputs and change the course of the conversation in a positive or negative way.
  • Army soldiers: USC Institute for Creative Technologies denotes a training simulation developed for Army soldiers that contains intelligent virtual agents.
  • Smartakus: A character developed at the German Research Center for Artificial Intelligence (DFKI) that functions as an anthropomorphic user interface. He can independently forward commands from the user to operate a device, and z. B. record a particular film with a DVD recorder.
  • Improv Puppets: These are virtual puppets (marionettes) that are used to interact with children. Through social contact with children, their cognitive development should be shaped.
  • Virtual Jack: Jack is an agent represented by a dog from the Petopia company. Jack is supposed to help customers navigate the website.
  • Erin the bartender: Erin O'Malley is a virtual bartender who can prepare and share hundreds of drinks. She also talks to the user about various topics.
  • Katherine, Interactive Newscaster: Katherine is a character on a website that provides the visitor with interactive news.
  • EACH Star Workshop: The EACH Star Workshop includes virtual teachers who can teach the visitor in the areas of management, social skills, foreign language, etc.
  • ERIC (Embodied Real-time Intelligent Commentary): is a rule-based framework with the help of which embodied graphical agents can be developed to comment on different events in real time. ERIC is currently being used to comment on a virtual horse race.

Core problems

The core problems in the conception and implementation of intelligent virtual agents lie on the one hand in the area of ​​graphic modeling and visualization, on the other hand in the real-time synchronization of speech, facial expressions and gestures. Another challenge is to make the agent as believable as possible.

Credibility in relation to virtual agents describes the willingness of the user to get involved in the illusion that arises from the interaction with the agent. As early as the early 1930s, Disney animators were making the audience laugh or cry by the way the characters expressed their emotions.

For a believable representation, the emotional state of a character, especially his goals, likes and dislikes, must be clearly defined and recognizable to the viewer at all times. In addition, emotions can be emphasized in order to give the viewer the time to grasp them.

Furthermore, the character's thought process has an impact on his feelings. The viewer must be able to understand the state in which the character is.

In practice, 7 characteristics have emerged that make a character appear realistic:

  1. talkative: The agent can credibly have a conversation. This includes z. B. that he takes the initiative to talk or can reflect different points of view.
  2. intelligent: The agent, for example, conducts a conversation within the framework that corresponds to his competence.
  3. individual: The agent has an independent personality and dynamic emotions, which he is currently expressing in his own way.
  4. social: The agent should have self-confidence in a social context.
  5. convincing: The agent should be able to react emotionally to events and situations and to respond to the emotions of his communication partner.
  6. Versatile: The agent should be random and changeable as to how, when and what his actions relate to.
  7. Consistent: The agent maintains basic elements of conversation, gestures, body language, and facial expressions in similar situations.

The implementation of virtual agents is an interdisciplinary process . Depending on the area of ​​application, not only graphic artists and AI researchers but also psychologists, sociologists and educators are involved in the development process.

Another problem with the implementation is the integration of many partly non-standardized software components that are used in different areas for development and lead to a complex system architecture.

The following sections describe other key issues in detail.

Visualization / graphic modeling and animation

Facial animation system "Alfred"

The graphic design and the animation of an intelligent virtual agent are the optical components that contribute to the credibility of the human user and leave a first impression on him.

In the beginning of the representation of virtual agents, mostly 2D objects were used. For example, context-dependent, animated GIF sequences can be combined into a whole range of motion animations. From this pool of animation fragments, the appropriate animation can then be selected during the interaction with a user at runtime and added to the one currently running.

Nowadays, 3D models are mainly used to represent virtual agents. The technical possibilities for creating graphics ( rendering (design) ) and animation in interactive mode are not comparable to those of current 3D animation films. Since the calculation time plays an essential role in the interaction with a user, no complex scenarios and detailed agents can be modeled. Details of the object to be modeled should, if possible, be processed in the textures and not in the 3D object in order to keep the total number of polygons as low as possible.

Well-known and widely used tools for 3D modeling would be for example:

  • Paid tools:
    • 3ds Max (3D Studio MAX)
    • Maya
    • Autodesk Softimage (XSI)
    • Cinema4D
  • Freeware:
    • Blender
    • AutoQ3D Community
    • Truespace

Behavior planning and behavior control

There are different approaches for behavior planning and control:

  • Scripting in computer games (e.g. LUA , which is used in games such as FarCry and World of Warcraft )
  • Plan-based systems (e.g. ABL, which was used in the development of Facade )
  • Rules-based systems (e.g. Jess, which was used to program ERIC's generic rule-based system, for example )
  • Cognitive architectures (e.g. SOAR , for which an example development is an educational agent named Steve )
  • Graph-based systems (e.g. SceneMaker).

With them the planning systems are created and solved in order to generate sequences of actions for the behavior of the intelligent virtual agent.

Basically, behavior planning and control can be divided into the areas of verbal (text and language-based communication) and non-verbal (facial expressions and gestures, but also e.g. movement in space) behavior.

Verbal behavior

The foundations for verbal interaction by and with intelligent virtual agents are speech recognition , speech processing and speech synthesis , with the help of which text and speech- based dialogue systems are generated. It is important to ensure that dialogues between humans and intelligent virtual agents appear as natural as possible.

The implementation can take place with the help of interpreters based on MarkUp languages ​​(e.g. VoiceXML ) or systems that have a graphical development environment (e.g. DialogOS ) and thus a very simple introduction to the creation of Offer dialogue systems.

Nonverbal behavior

The planning of non-verbal behavior includes in particular facial expressions, gestures and, in general, the agent's posture. In a broader sense, it also refers to general movements in space, which as pathfinding is a problem for AI. Animations for facial expressions and gestures, on the other hand, are described with the help of scripts. The animation is made up of 3 levels:

  1. Behavior planning
  2. Action planning
  3. Selecting and executing the animations needed to perform the actions.

Frameworks are available to help developers. One of the most widespread is the SAIBA framework, in which the behavior of the intelligent virtual agent is described with the aid of the Behavior Markup Language (BML). The BML script is then converted into an animation by a third-level engine (e.g. SmartBody, BML Realizer). A similar approach is chosen in EMBR, in which, however, an additional ( animation layer ) is introduced between the second and third level . This gives you more control over the animations themselves by first noting them with the help of a description language (EMBRScript) and only then converting them into the agent's concrete movement.

Technologies used

  • Microsoft Agents
  • Authoring tools, e.g. B. MASH
  • Standards
  • Markup languages ​​for specifying multimodal behavior

literature

  • Helmut Prendinger, Mitsuru Ishizuka (Ed.): Life-like characters. Tools, affective functions and applications. Berlin: Springer, 2004
  • Justine Cassell, Joseph Sullivan, Elizabeth Churchill, Scott Prevost: Embodied Conversational Agents , Cambridge: The MIT Pres, 2000, ISBN 0-262-03278-3 ( online )

Web links

References and comments

  1. ↑ Deviating: Software rendering in the English language Wikipedia
  2. ↑ Embodied Agent in the English language Wikipedia
  3. ↑ Digital Pet in the English language Wikipedia
  4. ^ Project Milo
  5. ↑ Façade interactivestory.net
  6. ^ USC Institute for Creative Technologies
  7. ^ German Research Center for Artificial Intelligence
  8. ↑ a b c d e f B. Hayes-Roth: What makes characters seem life-like? In: Life-like Characters. Tools, Affective Functions and Applications . Springer, 2003, ISBN 978-3-540-00867-5
  9. ↑ ERIC ( Memento of the original from February 17, 2011 in the Internet Archive ) Info: The archive link was automatically inserted and not yet checked. Please check the original and archive link according to the instructions and then remove this notice. the realtime commentary @1@ 2Template: Webachiv / IABot / www.martinstrauss.id.au
  10. ↑ Facial Animation System "Alfred" ( Memento of the original from November 19, 2010 in the Internet Archive ) Info: The archive link was inserted automatically and has not yet been checked. Please check the original and archive link according to the instructions and then remove this notice. @1@ 2Template: Webachiv / IABot / mm-werkstatt.informatik.uni-augsburg.de
  11. ↑ Pieter Spronck, Ida Sprink Huizen-Kuyper, Eric Postma: Enhancing the Performance of Dynamic Scripting in Computer Games In: Entertainment Computing - ICEC 2004 . Springer, 2004, ISBN 978-3-540-22947-6
  12. ↑ Michael Mateas, Andrew Stern: A Behavior Language: Joint action and behavioral idioms , 2004 ( PDF )
  13. ↑ Michael Mateas, Andrew Stern: Facade: An Experiment in Building a Fully-Realized Interactive Drama , 2003 ( PDF )
  14. ↑ Jess in the English language Wikipedia
  15. ↑ Michael Kipp, Martin Strauss: ERIC: a generic rule-based framework for an affective embodied commentary agent In: Proceedings of the 7th international joint conference on Autonomous agents and multiagent systems - Volume 1 , International Foundation for Autonomous Agents and Multiagent Systems, 2008 , ISBN 978-0-9817381-0-9 . Pp. 97-104
  16. ^ W. Lewis Johnson Jeff Rickel: Steve: an animated pedagogical agent for procedural training in virtual environments In: ACM SIGART Bulletin , ACM, 1997, ISSN  0163-5719 . Pp. 16-21
  17. ↑ B. Hartmann, M. Mancini, C. Pelachaud: Implementing expressive gesture synthesis for embodied conversational agents. In: Proc. of GW-2005 , 2005
  18. ^ S. Kopp, I. Wachsmuth: Synthesizing multimodal utterances for conversational agents. In: Computer Animation and Virtual Worlds 15 , 2004. pp. 39-52
  19. ↑ SAIBA
  20. ↑ BML
  21. ↑ SmartBody
  22. ↑ M. Thiebaux, S. Marsella, AN Marshall, M. Kallmann: Smartbody: Behavior realization for embodied conversational agents. In: Proc. of AAMAS-2008 , 2008. pp. 151-158
  23. ^ BML Realizer
  24. ↑ Alexis Heloir, Michael Kipp: EMBR - A Realtime Animation Engine for Interactive Embodied Agents In: IVA 2009 , Springer-Verlag Berlin Heidelberg, 2009. pp. 393–404