Scalable Adaptive Serious Games using Agent Organizations

Joost Westra,Frank Dignum,Virginia Dignum

[email protected]

Scalable Adaptive Serious Games using Agent Organizations

Overview

Introduction Adaptation to the trainee Organized adaptation of agents Scalability Conclusions

Serious Gaming

Dynamic Difficulty Adjustment

Online adaptation:Continuously balance challenges in the game with (developing) skills of the trainee

Current approaches

Fixed difficulties Central control or no coordination Mainly adjust simple subtasks

Agent Approach

Agents part of the design process Reasoning agents Adapting agents Specify boundaries of the adaptation (agent

organization)

Example:• Trainee is fire commander• 2 fireman agents• 1 victim agent• 1 agent controlling spreading of fires

Aspects

User• Evolving skills (when learning)

Characters • Characters adapt independently• Characters active for long periods, so, adaptation should

be believable Keep storyline

• Learning goals have to be maintained! Adaptation must be coordinated! Performance can not be measured separately for

each skill and influence of each agent

Story-line

Guarantee certain states are reached Subtasks defined by scene scripts and landmarks Connected by interaction structure

• Describes game progress• Connecting scenes• Tasks in parallel

• Start

• Get Access to Room

• Evacuate Victim

• Extinguish Fire

• End

Adaptation Engine

Coordinates task difficulty Check with game model Combinatorial auction

• User model• Agent preferences

Agent model• 2APL Agent

Agent model• Agent Bidding

• User Model

• Adaptation Engine

Update

Plans Bid

Task Weights Skill Levels

Selection

Preferences & Temination

Scene StatesApplicable plans

• Game Model

Start Get to site

Gather info

Secure area

Search building

Evacuate victims

Extinguish fire Clear area End

Agent Perspective

Agents Propose actions to adaptation engine at “natural” synchronization points

Created to facilitate trainee’s objectives (optimize agent behavior relative to trainee’s performance!)

Not responsible for suitable combination Conflict:

• Stay as consistent as possible• Propose enough actions

Adaptation engine can request agents to terminate behavior if necessary for coordination

Framework

Agent model• 2APL Agent

Agent model• Agent Bidding

• Game world

• Game state

• Agent interface

• User Model

• Adaptation Engine

• NPC

• NPC

• NPC

• NPC

Update

User Performance

Translate

Plans Bid

Update Beliefbase

Task Weights Skill Levels

External ActionSelectionGame Actions

Preferences & Temination

Scene StatesApplicable plans

• Game Model

Scalabilty: Scenes

Agents can only execute plans of active scenes Partial ordering gives a relatively low number of

concurrent scenes Sub-scenes: Even more fine grained pre-selection

Gather Info Search Building

Secure Area Evacuate Victims

Extinguish FireGet to site

Multiple victims

EndStart

Kitchen Fire

Scalabilty: Agent implementation

Active Subscenes are put in beliefbase

Only plans with active sub-scenes are applicable

Only plans for current goals are applicable

Other active beliefs can also restrict the number of applicable plans

Scalabilty: Believability

Agents estimate the Believabilty for each applicable action

Some actions clearly ruin the Believabilty of the agents • Believabilty 0 -> never suggest• Higher threshold than 0 is usually advisable -> even lower

number of suggestions Influence becomes bigger if the game progresses

• The player has more expectations on the behavior of the NPC

Scalabilty: Combination boundaries

Game model requirements can decrease the number of checked combinations

Influence greatly depend on the restriction• Easy:

– At least one fireman should perform X– Only one fireman available– Only evaluate combinations with the fireman performing X

• Difficult– X needs to be performed by at least two agents– No real indicator that other plans might not be better.

Scalability: Example

Assumptions• 30 different scenes

– 2 active at the same time• 4 subscenes

– 2 active at the same time• 6 plans per subscene per agent

Results• Naive:

– 720 active plans(30 scenes*4 sub-scenes*6 actions per sub-scene)• Agent Organization:

– 12 active plans (6 actions per sub-scene *2 sub-scenes active per scene * 2 active scenes /2 for believability filtering)

• 12.960.000 times as fast with only four agents

Scalability: Example, number of Agents

Conclusion

Continuous adaptation to the trainee Agent based approach

• Complex individual behavior and adaptation possible Agent organization for coordination

• Balance between individual flexibility and global story line maintaining learning goals

• Minimal central control for more efficiency and more flexibility

More scalable than centralized approach

[email protected]

Thank You!