Reactive Planning Agent Programming Language

1
Reactive Planning Agent Programming Language

• Duc Vo
• Decision Systems Laboratory
• School of Information Technology And Computer
  Science
• University of Wollongong

2
Index

• Introduction
• Review
• BDI Agents
• Available Agent Programming Languages
• Default Logics
• Reactive Planning Agent Programming Language
• Conclusion future research

3
Introduction (1)

• BDI Agents has been known as the best approach to
  model rational agents.
• There have several Agent Programming Language
  (APL) available
• 3APL (An Abstract Agent Programming Language)
• AgentSpeak(L)
• GOLOG (alGOl LOGic), ConGOLOG (Concurrent GOLOG)

4
Introduction (2)

• Working environment of rational agents is
  supposed to be open, dynamic, and ever-changing,
  where the agent can never have complete
  information about its environment
• Default logic introduced by Ray Reiter in 1981
  has been one of the most well known approach for
  reasoning when information is incomplete

5
Introduction (3)

• Agent programmed in available APLs is not able to
• Reasoning with incomplete information (e.g. have
  to wait until some information is discovered)
• Reconsidering a plan once the agent has been
  already executing it (e.g. when thing goes wrong,
  or thing turns out not as expected)

6
Motivating Example
Rescue Robot (RBot). Need to rescue a person at D
then carry him back to A. RBot can only travel on
available path and when the path is clear. RBot
can only see if a path to the next node is clear
or not Condition of out-of-sight path is unknown.
7
Review 3APL (1)

• 3APL An Abstract Agent Programming Language
• Developed by 3APL group at Institute of
  Information and Computing Sciences at University
  of Utrecht, Netherlands (http//www.cs.uu.nl/3apl)

8
Review BDI Agent

• Introduced by a philosopher Michael Bratman in
  1987
• Beliefs what agent knows about environment
• Desires what agent wants to achieve
• Intentions the plans that the agent intends to
  execute to achieve its desired goals

9
Review 3APL (2)
10
Review 3APL (3)

• PROGRAM "patrol_agent.3apl"
• CAPABILITIES
• at_east(self) WalkWest() NOT at_east(self),
  at_west(self),
• at_west(self) WalkEast() at_east(self), NOT
  at_west(self)
• BELIEFBASE
• at_west(self)
• GOALBASE
• patrol()
• RULEBASE
• patrol()  at_east(self)  WalkWest()
  patrol() ,
• patrol()  at_west(self) WalkEast() patrol()
  .

11
Review GOLOG (1)

• GOLOG alGO LOGic
• Where algo is an algebraic computer language
  for solving primarily mathematical and scientific
  problems using algorithms
• Developed by Cognitive Robotics Group at
  University of Toronto, Canada (http//www.cs.toron
  to.edu/cogrobo/)
• a logic programming language for agents

12
Review GOLOG (2)
13
Review GOLOG (3)

• PROGRAM "patrol_agent"
• Primitive actions WalkWest(), WalkEast()
• Fluents
• at_east(self, s) at situation s robot is at
  east
• at_west(self, s) at situation s robot is at
  west
• Primitive Action Pre Condition
• Poss(WalkWest(), s) ?? at_east(self, s)
• Poss(WalkEast(), s) ?? at_west(self, s)
• Successor State Actions
• Poss(a, s) ?
• (a WalkWest() and at_west(self, do(a,s)) or
• (a WalkEast() and at_east(self, do(a,s))

14
Review GOLOG (4)

• Procedures
• proc patrol(s)
• at_west(self, s)? WalkEast()
• or
• at_east(self, s)? WalkWest()
• endProc
• Initial Situation at_west(self, So)
• Run Program
• patrol(So)

15
Review AgentSpeak(L) (1)

• AgentSpeak(L) BDI Agents speak out in a logical
  computable language
• Developed by Anand S. Rao in 1996

16
Review AgentSpeak(L) (2)
17
Review AgentSpeak(L) (3)

• Plans
• !patrol at_east(self) ? WalkWest(), !patrol.
• !patrol at_west(self) ? WalkEast(), !patrol.
• Events
• lt!patrol, ltgt gt
• Beliefs
• at_east(self)
• Intentions
• lt !patrol at_east(self) ? WalkWest(),
  !patrol gt

18
Default Logic (1)

• Developed by Ray Reiter in 1981
• One of most well-known approach to non-monotonic
  reasoning
• Example 1
• Twitty is a bird (fact)
• Bird normally flies (default 1)
• Penguin normally does not fly (default 2)
• Twitty is a bird, and Twitty flies (extension)

19
Default Logic (2)

• Example 2
• Twitty is a bird (fact 1)
• Twitty is a penguin (fact 2)
• Bird normally flies (default 1)
• Penguin normally does not fly (default 2)
• Twitty is a penguin, Twitty is a bird, and Twitty
  flies (extension 1)
• Twitty is a penguin, Twitty is a bird, and Twitty
  does not fly (extension 2)

20
Reactive Planning APL

• When thing turns out not as expected or turns
  wrong, agent will search for alternative plans,
  or repairing existing intentions
• Background belief theory is used instead of just
  set of beliefs
• Preferred extension of background belief theory
  is affected by performed actions

21
Reactive Planning APL
22
Background Belief Theory
23
Background Belief Theory

• B (Delta, S, H, Ra)
• Where
• Delta is a default theory
• S is a selection function which selection an
  extension of default theory to be agent current
  set of beliefs
• H is an ordered set of performed actions (history
  of performed actions)
• Ra is a set of action-predicted-result rules
• A rule is of the form
• action conditions ? expected result

24
Plan repository

• Plan repository is set of pre-setup plans
• A plan is of the form
• trigger soft context, hard context ?
• actions, goals.
• Soft context needs satisfying when triggering the
  plan
• Hard context needs satisfying when triggering the
  plan or executing the plan
• A plan is called fail plan if its hard context
  fails during execution of the plan

25
Intentions

• An intention is an ordered stack of plans
• An intention is invalid if it has at least one
  fail plan
• Agent will only execute an intention if it is
  valid

26
Events

• Is of the form lttrigger, intentiongt
• Where trigger is either
• Add a goal
• Remove a goal
• Add a belief
• Remove a belief
• An event is an external event iff its intention
  is empty, it is an internal event otherwise
• An event is invalid iff its intention is invalid

27
Agent program

• Is of the form
• ltB, P, E, I, Sp, Se, Sigt
• Where
• B is a background belief theory
• P is a plan repository
• E is a set of events
• I is a set of intentions
• Sp is a selection function which selects an
  applicable plan for a trigger
• Se is a selection function which selects an event
  to process
• Si is a selection function which selects an
  intention to execute

28
Operational Procedures

• Event processing
• Intention executing
• Sensing procedure is a black-box in this program,
  however, result of sensing procedure or actions
  will create external events to set E of events

29
Event processing

• Begin Procedure
• select an event e
• get current agent belief set BS
• if (e is invalid) repair e end if
• if event trigger is to add a goal then
• select an applicable plan p
• create an intention i from p and intention of e
• add i to set I
• else if event trigger is to remove a goal then
• remove any intention to achieve that goal
• else if event trigger is to add a belief then
• revise background belief theory with this
  belief
• else if event trigger is to delete a belief then
• contract background belief theory with this
  belief
• end if
• Procedure

30
Intention executing

• Begin Procedure
• select an intention i
• get current agent belief set BS
• if (i is invalid) repair i end if
• if next element to be executed is an action then
• perform that action
• else if next element to be executed is a goal
  then
• create an internal event with i and that goal
  (add goal trigger)
• else if next element to be executed is a test
  belief then
• find the existential substitution from current
  set of beliefs for the test belief
• else if there is no more elements
• go to the next plan of i
• else if the i is empty
• remove i from I
• end if
• Procedure

31
Conclusion Future Research

• Agent program can
• Act proactively and reactively
• Reason with incomplete information
• Change beliefs subjectively (by means of
  selecting preferred extension of background
  belief theory and influent of history of action
  on this selection)
• and cannot
• Reason with time
• Distinguish between belief revision and belief
  update (e.g. when to update, when to revise)