ABSTRACT. We propose artificial game presenter avatars embodying affective .... from the avatar AI component; and (b) as a server of game hosting requests ...
Artificial Game Presenter Avatars Anthony Savidis1,2, Effie Karouzaki1 1
Institute of Computer Science, FORTH
2
Department of Computer Science, University of Crete {as,karuzaki}@ics.forth.gr
Enthusiasm, Positive surprise
Artificial Game Presenter Primary Game Terrain Roundtable Player Positions
Figure 1. The concept of computer games with artificial presenter avatars and its instantiation with our avatar. entertainment. Technically, all games played in such shows are multiplayer computer games with a technological setup amplifying social interaction. Essentially, the game presenter provokes social interaction to keep the players and the audience constantly motivated and alerted about the game progress. For this purpose, a show presenter relies on player profiles, current challenge and previous performance to provide feedback commonly involving humor, reward, sympathy, surprise, disappointment, enthusiasm, agony and anticipation. Clearly, presenters display an affective behavior. In this context, we have developed an artificial game-presenter avatar, named Amby, as a software system that can be incorporated in turn-taking multiplayer games. Amby was created to support the concept of multiplayer computer games hosted by artificial avatars as illustrated within Figure 1, top left. The top right part of Figure 1
ABSTRACT We propose artificial game presenter avatars embodying affective behavior to draw player-adapted social feedback during gameplay and introducing extra challenges to players called mini games, such as hangman and random card selection. The avatar’s AI was designed as an extension of the traditional sense-think-act loop of game characters to address the need for emotional reflection and adaptive reaction. We provide a cartoon-like 2d delivery for our avatar, however, one could support alternative approaches for rendering and animation.
1. INTRODUCTION Our work is motivated by the popularity of television game shows and the lack of an analogy in the domain of computer-based
l rna uli Stim
Adaptive Interaction
Enthusiasm
ns Encouragement / Discouragement
4.Adapt
1.Sense
eac
Game Characters
3.Think
rete R
Focus Area
Exte
Affective Computing
2.Reflect
Con c
Artificial Avatars
Happiness
Re ac tio
tion s
tiv fec Af
Ab str ac t
es tat eS
5.React
Positive / Negative Reassurance Anticipation Encouragement / Discouragement
Discouragement / Encouragement
Calm Anxiety
Anger
Negative / Positive Anticipation Discouragement / Encouragement
Disappointment Sadness
Performing Actions
Figure 2: The domain of our work (left), the enhanced behavior loop (middle) and the affective state space of our avatar (right).
Permission to make digital or hard copies of all or part of this work for personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior specific permission and/or a fee.
Permission to make digital or hard copies of part or all of this work or personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies bear this notice and the full citation on the first page. To copy otherwise, to republish, to post on servers, or to redistribute to lists, requires prior specific permission and/or a fee. $FH2009, 2FW -Oct , 2009, $WKHQV*UHHFH © ACM 2009 ISBN: 978-1-60558--/09/0...$10.00 415
shows how Amby has been incorporated within a pervasive board-game named Four Elements1 [2] with an overall physical setup inspired from game shows. To our knowledge there is no similar work proposing or implementing the idea of an artificial 1
A video of the Four Elements game is available from: http://www.ics.forth.gr/hci/files/plang/BoardGameVide(High%2 0res).wmv and a video of Amby in a play session is available from: http://www.ics.forth.gr/hci/files/plang/AmbyVideo.avi
presenter avatar for computer-games supporting affective behavior and player-adapted social reactions. Technically our work falls in the intersection of the four domains depicted under Figure 2, left.
4. SUMMARY Our work is motivated by the huge popularity of television game shows and the key role of the presenter towards their success. For this purpose we introduced the notion of an artificial gamepresenter avatar and we identified affective adaptive behavior to be the most important feature towards this direction. Clearly, anthropomorphism and quality of representation are other highlycritical dimensions that we do not address in the context of our work. To accomplish affective behavior we have implemented a scalable architecture relying on adaptation-processing loops. To our knowledge no earlier work proposed computer-games as socially-stimulating experiences motivating social peers to watch play sessions as in typical television game shows. In making the artificial intelligence core of our avatar we adopted a decision specification language we developed earlier named DMSL [3], while we have embedded its interpreter as part of the avatar implementation system.
2. BEHAVIOR The behavior loop of our avatar is an improvement of the traditional sense-think-act loop (see Figure 2, middle) for game characters introducing reflect and adapt as two extra processing stages. We logically split avatar reactions in two levels: (a) abstract reactions, denoting the general reaction category; and (b) concrete reactions, being alternative ways of specializing abstract reactions. Comparing to the Circumplex model [1] we dropped a few aspects (see being meaningless in a game context: (i) ‘fear’ and ‘disgust’ emotions; (ii) distinction among ‘activation’ and ‘deactivation’ (i.e. no tangible trophy); and (iii) intermediate states for sadness like ‘guilt’ or ‘depression’ (see Figure 2, right). The reflect process is implemented as a hybrid state transition network with transitions involving condition expressions (predicates). Player emotions relate to distinct affective states thus the transition logic is affect computation. In practice, emotional changes heavily depend on the summative effect of other game incidents (events). For this purpose we introduced summative variables that can be involved in transition predicates, together with memory variables, normally recording game events.
The primary extensions we introduced to the basic sense-think-act behavior processing concerned: (a) a reflect stage to derive players emotions (affective processing); and (b) an adapt stage to specialize avatar reactions in ways bets-fitting individual player profiles. An extra feature we included is the on-the-fly generation of a textual script in the form of dialogues amongst players and the avatar. As part of our future extensions we work on the structuring of rule libraries in ways enabling meta-rules choose alternative avatar behavior styles such as humorous, ironic, teasing or compassionate.
3. ARCHITECTURE
Avatar Use r Inte rfac e [ R e nd e r in g , A n im a tio n, Sp e e c h ,]
Im a ge L ib ra r y
Text L ib r a ry
A u d io Li br a r y
sense reflect think adapt react
audio / music expression side action
inventory
Game
players mini games lives/score/dice
Ex p r e s s io n L ib ra ry
Protocol-1
M ana gem e nt of su m m ativ e and m em or y va riable s
A vatar Proce ssing a nd Coordination [ M ain Loo p ]
Game Monitoring Client Library
U tility F unc tions Ex porte d to D M S L
comment
Avatar User Interface
DM SL libra ry (interpreter a nd pro file m ana ger )
AI
Protocol-2
P la y e r p r o fil e s
protocol-3
Adap t r ule s
[1] J. Russell, G. Lemay (2000). Emotion Concepts. In Handbook of Emotion, M. Lewis, M. Haviland-Jones (Eds.), New York: Guilford Press. [2] A. Savidis, Y. Lilis (2008). Adaptable pluggable multimodal input with extensible accessible soft dialogues for games. In ACM ACE 2008 Int. Conf. on Advances in Computer Entertainment, pp. 155-158. [3] A. Savidis, M. Antona, C. Stephanidis (2005). A Decisionmaking Specification Language for Verifiable User-interface Adaptation Logic. Journal of Software Engineering and Knowledge Engineering, Volume 15, Issue 6 (December 2005), pp. 1063-1094.
Game Hosting
T h in k ru le s
5. REFERENCES
Presenter Personality
R e fle c t ru le s
protocol-3
The overall architecture is illustrated under Figure 3, left part, showing the split amongst the decision rules categories (such as think and reflect), as well as the various external libraries deployed encompassing data required for the implementation of the avatar physical reactions (such as images, audio files and expression animations). Also, the dual role of the avatar User Interface is outlined: (a) as a server of “reaction” requests coming from the avatar AI component; and (b) as a server of game hosting requests coming directly from the game to support inventory, player display, turn taking functionality, and mini games. All communication between the avatar and the AI module or the game core takes places over the network custom protocols. A couple of scenes from hangman game sessions are shown under Figure 3, right part, illustrating the respective emotion states.
Figure 3. Software architecture of our avatar (left), and details of the User Interface component architecture (right).
416
