An Overview of Goals and Goal Selection

1
An Overview of Goals and Goal Selection

• Justin L. Blount
• Knowledge Representation Lab
• Texas Tech University
• August 24, 2007

2
Outline

• Goals
• Goals in ASP agents
• Goals in Situation Calculus agents
• Goals in BDI agents

3
Goals

• What question do I want to answer
• What do I do now? (goal/planning)
• What do I want? (goal selection)
• How do I get it? (planning)
• What is a goal?
• How to choose/select a goal?
• Goal n. the result or achievement toward which
  effort is directed aim end. (dictionary)

4
Goals in ASP agents (Baral, Gelfond, 2000)

• Assume at each moment t the agents memory
  contains the domain description and a partially
  ordered set G of the agents goals.
• A goal is a finite set of fluent literals the
  agents wants to make true.
• Partial ordering corresponds to the comparative
  importance
• Agent loop
• 1. Observe world
• Select one of the most import goal g in G to be
  achieved
• Find plan a1, an to achieve g
• 4 Execute a1

5
Goals in ASP agents (Balduccini, 2005)

• Agent loop
• 1. Observe world
• Select a goal
• Find plan a1, an to achieve g
• 4 Execute a1
• The selection of the current goal is performed
  taking into
• account information such as the partial ordering
  of goals, the
• history of the domain, the previous goal, and
  the action
• description (e.g., to evaluate how
  hard/time-consuming it
• will be to achieve a goal).

6
Goals in Situation Calculus agents (Shapiro,
Lesperance, 2001)

• Consistent set of goals --If the agent gets a
  request for g and it already has the goal that -g
  , then it does not adopt the goal that ,
  otherwise its goal state would become
  inconsistent and it would want everything.
• Paths to goal are finite. A maintenance goal of
  X is always true is not finite, but can do X is
  true for next 100 time steps

7
Goals in Situation Calculus agents (Shapiro,
Lesperance, 2005)

• Expansion
• An agents goal are expanded when it is requested
  to do something by another agent. Unless it
  currently has a contradicting goal.
• Contraction
• If an agents owner REQUESTS(x) then later
  changes his mind. The owner uses a
  CANCEL_REQUEST(x). Can only be used if a REQUEST
  was executed previously
• Persistence
• A goal x persists over an action a, if a
  is not CANCEL REQUEST, and the agent knows that
  if x holds then a does not change its value.

8
Goals in situation calculus (Sardina, Shapiro,
2003)

• Prioritized goals. Each goal has a priority
  level
• an agent that will attempt to achieve as many
  goals as possible in priority order even if the
  agent does not know of a plan that is guaranteed
  to achieve all the goals.
• Priorities are strict
• A strategy to achieve 1 High level goal is
  preffered to strategy to achieve many (or all)
  lower level goals

9
Goals in Situation Calculus agents
(Shapiro,Lesperance, 2007)

• An agent should drop goal that it believes are
  impossible to achieve.
• However, if the agent revises its beliefs, it may
  later come to believe that it was mistaken about
  the impossibility of achieving the goal. In that
  case, the agent should readopt the goal.
• If an agent receives a request to adopt goal X,
  it will adopt it if it does not conflict with a
  higher priority goal.

10
Goals in Situation Calculus agents (Shapiro,
Lesperance, 2007)

• Goal should be compatible with beliefs. The
  situations that the agent wants to actualize
  should be on a path from a situation that the
  agent considers possible.
• Instead of checking whether each individual goal
  is consistent with beliefs, check if the set of
  all goals are consistent with beliefs
• it could be the case that each goal is
  individually compatible with an agents beliefs
  but the set of goals of the agent is
  incompatible, so some of them should be dropped.
• Which ones should be dropped?
• Each agent has a preorder over goal formulae that
  corresponds to a prioritization of goals
• Chooses a maximal subset respecting this ordering

11
Goals in BDI agents (DIverno,Kinny,Luck,Wooldridg
e,1998)

• Goals correspond to the tasks allocated to it
• From their agent loop
• generate new possible desires (tasks), by finding
  plans whose trigger event matches an event in the
  event queue
• A plan consists of subgoals or primitive actions
• Thus an agent with goal achieve PHI has a goal
  of performing some (possibly empty) sequence of
  actions, such that after these actions are
  performed, PHI will be true.
• Thus an agent with goal query PHI has a goal
  of performing some (possibly empty) sequence of
  actions, such that after it performs these
  actions, it will know whether or not PHI is true.
• Thus an agent can have a goal either of achieving
  a state of affairs or of determining whether the
  state of affairs holds.

12
Goals in BDI agents (Thanagarajah, Padgham,
Harland, 2002)

• Desires may be inconsistent
• Goals must be consistent
• if it is not possible to immediately form an
  intention towards a goal then the goal is simply
  dropped.
• It certainly seems more reasonable that the agent
  have the ability to remember a goal, and to
  form an intention regarding how to achieve it
  when the environment is conducive to doing so.
• How to choose between two mutually inconsistent
  goals?
• if a new goal X is more important than an
  existing goal Y with which it conflicts, then Y
  should be aborted and pursued. Otherwise, (X is
  less important or same importance as Y ), X is
  not adopted.
• (too nieve)
• if there is no preference ordering between two
  goals, then we should prefer a goal that is
  already adopted over one that is not

13
Goals in BDI agents (Winikoff, Padgham,Harland,
2001)

• Problem - BDI is difficult to explain and teach
• Solution - simplify
• Explicitly represent goals. (instead of desires)
• This is vital in order to enable selection
  between competing goals, dealing with conflicting
  goals, and correctly handling goals which cannot
  be pursued at the time they are created and must
  be delayed.
• Highlight goal selection as an important issue.
  By contrast, BDI systems simply assume the
  existence of a selection function.
• avoidance goals, or safety constraints
• (e.g. never move the table while the robot is
  drilling).

14
Goals in BDI agents(Winikoff, Padgham, Harland,
Thangarajah,2002)

• Goals have 2 aspects-- declarative and procedural
• Declarative -- to reason about important
  properties of goals
• Procedural -- to ensure goals can be achieved
  efficiently in dynamic environments
• Reasoning about multiple goals (simple case 2
  goals)
• Plans to achieve both may be independent
• irrational to try to achieve both X and -X
  simultaneously
• Necessarily consistent -- iff all possible
  subgoals do not conflict
• Necessarily inconsistent -- iff some necessary
  subgoals conflict
• Possibly inconsistent -- choose a consistent
  means on achieving both
• Necessarily support -- both share a common
  necessary subgoal
• Possibly support -- the exists a common necessary
  subgoal

15
Goals in BDI agents(Pohkar, Braubach,
lamersdorf,2005)

• Goal types -- perform, achieve, query, maintain.
  
• Active goals are currently pursued
• Options are inactive because the agent explicitly
  wants them to be
• Ex the option conflicts with a active goal
• Suspended goals must not be pursued because their
  context is invalid
• Will remain inactive until their context is valid
  and they become options
• Deliberation is executed when one of the
  following occurs
• Creation condition -- defines when new goal
  instance is created
• Context condition -- defines when a goals
  execution should be suspended
• Drop condition -- defines when a goal instance is
  removed
• Inhibition arc -- define a negative relationship
  between 2 goals
• used in deliberation, constrain what goals are
  reconsidered

16
Goals in BDI agents(Duff, Harland,
Thangarajah,2006)

• Maintenance goals - defines states that must
  remain true rather than a state that is to be
  achieved.
• Reactive - goals are only acted upon when the
• maintenance condition is no longer true.
• Proactive - anticipate failures and act in order
  to prevent them from failing
• ( done by performing actions that we prevent
  the condition from failing or suspending goals
  that will cause the maintenance condition to
  fail)
• Future -- prioritize maintenance goals via urgency

17
Goals in BDI agents(Morreale,et al 2006)

• A goal g1 is inconsistent with a goal g2 if and
  only if when
• g1 succeeds, then g2 fails.
• agent deliberates and generates g as an option
• agent checks if g is possible and not
  inconsistent with active goals
• If both checks are passed then g becomes and
  intention
• If case of inconsistency among g and some active
  goals
• g becomes intention only if it is prefferred to
  such inconsistent goals which will be dropped
• Preference relation -- not total
• since several goals can be pursued in parallel,
  there is no need to
• prefer some goal to another goal if they are not
  inconsistent each
• other.

18
References

• 1 Baral,C. and Gelfond, M. 2000. Reasoning
  agents in Dynamic Domains, Logic Based
  Artificial Intelligence , Edited By J. Minker,
  Kluwer.
• 2 Balduccini, M. 2005. Answer Set Based Design
  of Highly Autonomous, Rational Agents. PhD
  thesis, Texas Tech University.
• 4 Shapiro, S. and Lesperance, Y. 2001.
  Modeling multiagent systems with the cognitive
  agents specification language a feature
  interaction resolution application. In C.
  Castelfranchi and Y. Lesperance, editors, Proc.
  ATAL-2000, pages 244259. Springer-Verlag,
  Berlin.
• 5 Sardina, S. and Shapiro, S. 2003 Rational
  action in agent programs with prioritized
  goals.AAMAS, 417-424
• 6 Shapiro, S., Lesperance, Y., and Levesque,
  H., 2005. Goal Change, in Proccedings of the
  IJCAI-05 Conference, Edinburgh, Scotland.
• 7 Shapiro, S. and Brewka, G. 2007. Dynamic
  Interactions between Goals and Beliefs. IJCAI,
  2625-2630
• 8 dInverno, M. Kinny, D. Luck, M. and
  Wooldridge,M. 1998. A Formal Specification of
  dMARS, In Intelligent Agents IV In Proceedings of
  the Fourth International Workshop on Agent
  Theories, Architectures and Languages, Singh, Rao
  and Wooldridge (eds.), Lecture Notes in
  Artificial Intelligence, 1365, Springer-Verlag.

19
References - continued

• 9 Thangarajah, J., Padgham, L., and Harland,
  J. 2002. Representation and reasoning for goals
  in BDI agents. In Proceedings of the Twenty-Fifth
  Australasian Computer Science Conference (ACSC
  2002), Melbourne, Australia.
• 10 Winikoff, M., Padgham, L., and Harland, J.
  2001. Simplifying the Development of Intelligent
  Agents. In AI2001 Advances in Artificial
  Intelligence. 14th Australian Joint Conference on
  Artificial Intelligence. LNAI 2256, pages
  557-568, Adelaide.
• 11 Winikoff, M., Padgham, L., Harland, J., and
  Thangarajah, J. 2002. Declarative and Procedural
  Goals in Intelligent Agent Systems, Proceedings
  of the Eighth International Conference on
  Principles of Knowledge Representation and
  Toulouse.
• 12 Pokahr, A., Braubach, L., Lamersdorf, W.
  2005. A Goal Deliberation Strategy for BDI Agent
  Systems, Third German conference on Multi-Agent
  System Technologies
• 13Duff, S., Harland, J., and Thangarajah, J.
  2006. On Proactivity and Maintenance Goals,
  Proceedings of the Fifth International Conference
  on Autonomous Agents and Multi-Agent Systems
  (AAMAS'06), Hakodate.
• 14Morreale, V., Bonura, S., Francaviglia, G.,
  Centineo, F., Cossentino, M., and Gaglio, S.
  2006. Reasoning about Goals in BDI Agents the
  PRACTIONIST Framework. Proc. Of the Workshop on
  Objects and Agents.  Catania, Italy.

20
(No Transcript)