Blackboard Architectures

1
Blackboard Architectures

• Michael Zyda
• Zyda_at_isi.edu

2
A hierarchy of motion behaviors
3

• So the next thing I am going to discuss is how we
  can use a Blackboard Architecture for this
  hierarchy
• Paper to read
• Damian Isla Bruce Blumberg Blackboard
  Architectures, AI Game Programming Wisdom,
  Volume 1, pp. 333 - 344.
• These notes are built from that paper as enhanced
  by some of my own thoughts

4
Blackboard architectures

• Any type of social system requires some
  coordination of action.
• Armies, bees, wolves, sports team
• Individuals need to take certain roles, and
  high-level goals need to be served by
  occasionally non-obvious, low-level cooperative
  action.
• This is as true for groups of simulated agents in
  a computer game as it is for natural and social
  systems.
• As the number of agents in a system increases
  along with the number of potential roles for
  individual agents, controlling coordinating
  their behavior becomes increasingly difficult.

5

• The blackboard approach is one possible means of
  handling this coordination.
• Although simple to implement, the architecture
  has proven elegant and powerful enough to be
  useful for problems ranging from synthetic
  character control to natural language
  understanding and other reasoing problems.
• Blackboards originated as a technique for formal
  reasoning and problem solving in the 1970s.

6
Metaphor

• The blackboard is built around a metaphor the
  physical blackboard we all used at school wa
  sitself a problem solving tool.
• We have all experienced crowding around a
  blackboard with a group of friends or colleagues
  to tackle a particularly nasty problem.
• In these cases, the blackboard was useful because
  it provided a shared space in which the problem
  could be broken down and incrementally solved.
• So think about your friends standing around the
  blackboard each willing to chip in to help solve
  the global problem you all are working on

7
(No Transcript)
8
Blackboard

• Blackboard
• A publicly readable/writable information display.
• Typically, the blackboard contains a list of
  logical assertions upon which other components
  operate.
• The blackboard is usually organized in some
  fashion so that the external components (agents)
  need only explore specific areas of the
  blackboard for the contents in which they are
  interested.
• Some systems, for example, impose a hierarchy to
  the assertions, identifying some assertions as
  data (the initial input to the system), and other
  assertions as goals, with a specified number of
  intermediate levels.
• Some systems also annotate assertions with a
  credibility rating, indicating how likely the
  assertion is.

9
Knowledge Sources (KSs)

• Knowledge Sources (KSs) -
• Flanking the blackboard are a series of
  components that are able to operate on the
  information that the blackboard contains.
• These are the specialists (agents) that will
  collaborate to solve the problem.
• Like all specialists, KSs only have very narrow
  regions of expertise, and so only know what to do
  in a very narrow set of circumstances.
• Usually, KSs are inactive, awaiting specific sets
  of preconditions to become true.
• A KS might also return a willingness to run, or
  relevance, indicating its degree of applicability
  to the current contents of the blackboard.

10

• Knowledge Sources (KSs) -
• When KSs are found to be relevant, they can be
  executed, their actions modifying the contents of
  the blackboard.
• These modifications might take the form of new
  logical assertions, changes to existing
  assertions, or control signals to other KSs.
• Significantly, KSs are only allowed to
  communicate with one another through the
  blackboard.

11
Arbiter

• Arbiter -
• Given a single snapshot of the blackboard
  contents, it is possible that any number of KSs
  can indicate a nonzero relevance.
• It is the job of the arbiter to decide which of
  the relevant KSs to execute.
• This is a critical step, and constitutes the
  entirety of the control strategy for the entire
  blackboard system.
• In many of the earlier systems, the arbiter would
  pick a single winning KS - this was a means of
  ensuring that two conflicting actions were not
  taken in a single timestep.
• A trivial strategy for picking a single winner
  would simply be to pick the KS with the highest
  self-generated relevance.

12

• Arbiter -
• However, any number of strategies is possible,
  including ones that incorporate a focus of
  attention, a motivational state, an emotional
  state, a personality model, and so forth.
• Such state information is additional input to the
  arbiter to help it make its decision.
• An advanced arbiter might choose the KS based not
  only on immediate relevance (which is what the KS
  returns), but also on expected relevance to the
  overall goal, resulting in behavior that is both
  data-driven and goal-driven.

13

• On a single update, a list of relevant KSs is
  collected based on recent changes to the
  blackboard.
• This list is passed to the arbiter, which uses
  whatever methods it wants to choose a winning KS.
• This KS is then executed, causing further changes
  to the blackboard.
• For some problems, the new state is examined for
  termination conditions that would indicate that
  the problem is solved (I.e. that the blackboard
  contains the solution).
• If the termination conditions are not met, then
  the cycle is repeated.

14

• Difference between a rule-based system and a
  blackboard -
• Whereas many rule-based systems typically require
  a compact and uniform encoding of both
  trigger-conditions and actions in their rules,
  blackboards allow arbitrary code to be executed
  for both these parts.
• Where one KS (the blackboard rule equivalent)
  might act solely based on the contents of the
  blackboard, another might pause execution and
  solicit input from the user, or perform any
  arbitrary complex operation.
• A single KS could encapsulate an entire subsystem.

15

• Difference between a rule-based system and a
  blackboard -
• Whereas classical rule-chaining systems allow
  only forward-chaining or only backward-chaining,
  blackboard systems impose few enough constraints
  to allow one or the other or both at once.
• Related to the above, blackboards allow multiple
  concurrent lines of reasoning.
• There might be multiple strategies for solving a
  single problem, for example, and both might be
  investigated at the same time.
• Maintaining these lines of reasoning depends, of
  course, on the sophistication of the arbiter.
• Note that in principle, nothing prevents two
  contradictory assertions from being present in
  the blackboard at the same time.
• It is simply assumed that the structure of the
  blackboard will eventually reconcile the
  contradiction at a higher level, perhaps by
  discarding the assertion with the lowest
  credibility.

16
Control rather then reasoning

• One important trend is that blackboards are being
  used increasingly for control rather than for
  reasoning.
• Although the line between the two is often hazt,
  it might generally be considered that the weight
  of decision-making has been moved out of the
  blackboard and arbiter and into the KSs.
• As stated already, a single KS might encapsulate
  an entire subsystem that uses the blackboard
  simply as the medium through which it issues and
  receives instructions from other subsystems.

17

• The assertions of the blackboard in this case are
  actually control signals.
• Despite these differences, two fundamental
  aspects of the blackboard remain
• A KS need not know (or need not constrain) when
  and how the assertions or control signals it
  produces will be used.
• A KS need not know (or need not heed) the
  originator of the assertions or control signals
  it acts upon.

18
(No Transcript)
19
(No Transcript)
20
Clean Interfaces

• Clean interfaces
• In the end, a blackboard is a clean interface
  that all new threads can read/write in a uniform
  fashion.
• Multiple threads
• How does a thread (KS, agent) know when to come
  awake and put itself to asleep.

21
C4
22
BBWAR

• On the CD is the Java source for BBWar, a simple
  RTS-like game built around a blackboard
  architecture
• The sample code shows you how a hierarchy is
  achieved with a blackboard.
• Opportunistic
• Cooperative
• Coordinated

23
(No Transcript)