Blackboard Architectures

1
Blackboard Architectures

• Damian Isla and Bruce Blumberg

2
The brain architecture (C4)
3
Abstract

• As the number of agents increases, controlling
  and coordinating their behavior becomes difficult
• The blackboard approach
• Useful for synthetic character control, natural
  language understanding, reasoning problems,
• Divide and conquer

4
The Canonical Blackboard Architecture 1/2
5
The Canonical Blackboard Architecture 2/2

• Blackboard
• A publicly read / writable information display
• KSs (Knowledge Sources)
• Specialists with relevance
• Operate on the information that the blackboard
  contains
• Only communicate each other through the
  blackboard
• Arbiter
• Decide the relevant KSs to execute
• Data-driven vs. Goal-driven

6
Rule-based vs. Blackboard

• Rule-based system
• Compact and uniform encoding of both conditions
  and actions
• Allow only forward-chaining or only
  backward-chaining
• Blackboard
• Allow arbitrary code to be executed
• Allow one, the other, or both at once
• Allow multiple concurrent lines of reasoning

7
The trend of blackboard

• Using for control rather than reasoning
• Decision-making is moving out of blackboard and
  arbiter into the KSs
• Two fundamental aspects
• A KS needs not know
• when and how the assertions or control signals it
  produces will be used
• the originator of the assertions or control
  signals it acts upon

8
Blackboards for Intra-Agent Coordination 1/2

• Two alternative but equivalent schemes
• The advantage of the blackboard structure
• Scalability
• Encapsulation

9
Blackboards for Intra-Agent Coordination 2/2
10
Blackboards for Inter-Agent Coordination 1/3

• Coordinate the activity of multiple agents
• BBWar
• KSs
• Individual military units that have special
  abilities
• Each unit has a set of skills (action)
• Blackboard
• Take the form of open missions
• Skill name, a priority level, a capacity,
  arbitrary data

11
Blackboards for Inter-Agent Coordination 2/3
12
Blackboards for Inter-Agent Coordination 3/3

• Implementation

ActionSelect(time, unit, blackboard) Skills ß
getSkills(unit) For each element, c, of list
skills List rel-list blackboard.read(skillsc)
rel_missions ß Concatenate(rel_missions,
rel-list) For each element, c, of
rel_missions relevancesc CalculatePriority(uni
t, rel_missionsc) w_index highest(relevances
) if (relevancesw_index gt CalculatePriority(uni
t, curMission)) ApplyForMission(unit,
rel_missionsw_index, relevancesw_index)
13
Conclusions

• The blackboard architecture constrains that KSs
  communicate only through the blackboard
• Opportunistic, cooperative, coordinated, and
  extensible behavior
• Provide a good framework around which to
  structure and think about these tasks