Trabajo Regular presentado en CoSECiVi'14. Resumen: Authoring believable Non-Player Characters (NPCs) - which is a bottleneck in the development process - and generating interesting and varied stories where they interact - which is a key aspect to enrich the user's experience - are major concerns in Interactive Storytelling (IS). In this paper we propose an authoring framework that separates the autonomous behaviors of each NPC (that is, its basic capabilities and their feasible combinations) from the coordination regimes that drive the interactions among characters, thus facilitating the reuse of knowledge in order to reduce the authoring effort. A decentralized multi-agent narrative engine is also presented, which uses this encapsulated knowledge to provide story variability, simulating several actors and a performance director who collaborate to build the story. An implementation using Hierarchical Task Networks (HTN) planning is described.

1. SEPARATING THE AUTONOMOUS BEHAVIORS
AND COORDINATION REGIMES OF
NON-PLAYER CHARACTERS
Gonzalo Milla-Millán and Juan Fdez-Olivares
gmillamillan@decsai.ugr.es
Department of Computer Science and
Artificial Intelligence, CITIC-UGR, University
of Granada, Spain

2. A Motivating Example
 A motivating example (extracted from [1]):
 Jerry and George want to leave a room, and so they
independently walk towards the door, which can only fit
one person through at a time. Jerry graciously permits
George to leave first.

3. Overview
 Goals:
 Reduce the authoring effort in Interactive Storytelling (IS)
systems (bottleneck in the development process)
 Provide story variability (improves user’s experience)
 Means:
 Separate the autonomous behaviors of NPCs from their
coordinations regimes
 Authoring framework to encapsulate the knowledge and allow for
its reuse
 Narrative engine to use the encapsulated knowledge in order to
automatically generate varied stories
 Implementation using Hierarchical Task Network (HTN)
planning

4. Authoring Framework
 Physical Aspects of the virtual world
 Context Model (a room with a door, positions of the NPCs, etc.)
 Autonomous NPCs present in the world (Jerry and George)
 Basic capabilities performable by NPCs that can alter the world (walk-to-door, cross-door)
 Goal-oriented and context-dependent autonomous behaviors ( Leave-room : (walk-door  cross-door) || (cross-door) )
 Coordination regimes
 Interdependencies (require-door-available / not-require-door-available)
 Operational Relationships ( (Jerry < George) || (George < Jerry) )

5. Authoring Framework
 The physical aspects of the virtual world are common
 Behavior profiles and coordination regimes are encoded separatedly, relying on
the common physical aspects of the virtual world.
 Behavior profiles are collections of goal-oriented autonomous behaviors, which can
be organized in libraries and assigned to NPCs.
 Coordination Regimes are encoded separatedly from the autonomous behaviors
and can also be grouped in libraries.

6. Decentralized Narrative Engine
 n Actor agents + 1 Performance Director (Actor != NPC)
1. The Performance Director (PD) sends to each actor her narrative goal (e.g., leave room)
2. Each Actor (A) sends back to the PD her intended autonomous behavior (e.g., walk-door  cross-
door)
3. The PD applies the Coordination Regime (e.g., require-door-available + (Jerry < George)) to the
set of intended autonomous behaviors proposed by each actor
4. The output is a coordinated story

7. Authoring using HTN planning

8. Authoring using HTN planning

9. Authoring using HTN planning
Door
availability
is NOT
managed

10. Authoring using HTN planning

11. Authoring using HTN planning
Door
availability
IS managed

12. Authoring using HTN planning
;; The virtual world is represented using ground predicates according
;; to the context model
(init:
(NPC-in-room jerry room1) (NPC-in-room george room1)…)
;; The narrative goals are represented using ground goal tasks
(:goal
(task-goal (Leave-room jerry)))

13. Narrative Generation using HTN
planning
 Examples of Coordinated Stories
1. ( ( [ (1 walk J D) (1 walk G D) ] ) (2 cross J D) (3 cross G D) )
2. ( ( [ (1 walk J D) (1 walk G D) ] ) (2 cross G D) (3 cross J D) )
3. ( ( [ (1 walk J D) (1 walk G D) ] ) ( [ (2 cross J D) (2 cross G D) ] ) )
World

14. Conclusion
 Separation of autonomous behaviors from coordination
regimes
 Reduce authoring effort (by allowing reuse of knowledge)
 Provide story variability
 Authoring framework
 Decentralized narrative engine
 Use of HTN planning by both the actors and
performance director agents ensures the compliance of
the generated narrative with the constraints imposed by
 The behavior profiles of the involved NPCs
 The applicable coordination regime

15. Future Work
 More complex behaviors for NPCs ([2])
 Consider more types of interdependencies ([1])
 Integration with a Game Engine (Unity3D)
 Integration of the player (continual planning approach:
execution + monitoring + repair/replan)
 Dynamic management of the applicable coordination
regime (operational relationships that evolve along
time, interdependencies managed according to an
evaluation function for the dramatic value of the
narrative, etc…)
 Different layers of knowledge for interaction (e.g.,
social relationships)

16. Bibliography
1. Smith, B. L.; Tamma, V. & Wooldridge, M. An ontology for
coordination. Applied Artificial Intelligence, Taylor &
Francis, 2011, 25, 235-265
2. Paul, R.; Charles, D.; McNeill, M. & McSherry, D. MIST: an
interactive storytelling system with variable character
behavior Interactive Storytelling, Springer, 2010, 4-15