introduction agent programming

Koen Hindriks Multi-Agent Systems

IntroductionAgent Programming

Koen HindriksDelft University of Technology, The Netherlands

Learning to program teaches you how to think.

Computer science is a liberal art.

Steve Jobs

Koen Hindriks Multi-Agent Systems 2012 2 Koen Hindriks Multi-Agent Systems 2

Outline

• Previous Lecture, last lecture on Prolog:– “Input & Output”– Negation as failure– Search

• Coming lectures:– Agents that use Prolog

• This lecture:– Agent Introduction– “Hello World” example in the GOAL agent

programming language


Agents: Act in environments

Choose an action

Percepts

Action

environment

agent


Agents: Act to achieve goals

Percepts

Action

events

actions goals

environment

agent


Agents: Represent environment

Percepts

Action

events

actions goals

plans

beliefs

environment

agent


Agent Oriented Programming• Agents provide a very effective way of building

applications for dynamic and complex environments

+• Develop agents based on Belief-Desire-Intention

agent metaphor, i.e. develop software components as if they have beliefs and goals, act to achieve these goals, and are able to interact with their environment and other agents.


A Brief History of AOP• 1990: AGENT-0 (Shoham)• 1993: PLACA (Thomas; AGENT-0 extension with plans)• 1996: AgentSpeak(L) (Rao; inspired by PRS)• 1996: Golog (Reiter, Levesque, Lesperance)• 1997: 3APL (Hindriks et al.)• 1998: ConGolog (Giacomo, Levesque, Lesperance)• 2000: JACK (Busetta, Howden, Ronnquist, Hodgson)• 2000: GOAL (Hindriks et al.)• 2000: CLAIM (Amal El FallahSeghrouchni)• 2002: Jason (Bordini, Hubner; implementation of AgentSpeak)• 2003: Jadex (Braubach, Pokahr, Lamersdorf)• 2008: 2APL (successor of 3APL)This overview is far from complete!


A Brief History of AOP• AGENT-0 Speech acts• PLACA Plans• AgentSpeak(L) Events/Intentions• Golog Action theories, logical specification• 3APL Practical reasoning rules• JACK Capabilities, Java-based• GOAL Declarative goals• CLAIM Mobile agents (within agent community)• Jason AgentSpeak + Communication• Jadex JADE + BDI• 2APL Modules, PG-rules, …


Outline

• Some of the more actively being developed APLs

– 2APL (Utrecht, Netherlands)

– Agent Factory (Dublin, Ireland)

– GOAL (Delft, Netherlands)

– Jason (Porto Alegre, Brasil)

– Jadex (Hamburg, Germany)

– JACK (Melbourne, Australia)

– JIAC (Berlin, Germany)

• References


2APL – Features2APL is a rule-based language for programming BDI agents:•actions: belief updates, send, adopt, drop, external actions•beliefs: represent the agent’s beliefs•goals: represents what the agent wants•plans: sequence, while, if then•PG-rules: goal handling rules•PC-rules: event handling rules•PR-rules: plan repair rules


2APL – Code SnippetBeliefs: worker(w1), worker(w2), worker(w3)

Goals: findGold() and haveGold()

Plans: = { send( w3, play(explorer) ); }

Rules = {

… goal handling rule

G( findGold() ) <- B( -gold(_) && worker(A) && -assigned(_, A) ) |

send( A, play(explorer) );

ModOwnBel( assigned(_, A) );

,

E( receive( A, gold(POS) ) ) | B( worker(A) ) -> event handling rule

{ ModOwnBel( gold(POS) );

},

E( receive( A, done(POS) ) ) | B( worker(A) ) -> explicit operator for events

{ ModOwnBel( -assigned(POS, A), -gold(POS) );

},

…

}modules to combine and structure rules


JACK – FeaturesThe JACK agent Language is built on top of and extends Java and provides the following features:•agents: used to define the overall behavior of mas•beliefset: represents an agent’s beliefs•view: allows to perform queries on belief sets•capability: reusable functional component made up of plans, events, belief sets and other capabilities•plan: instructions the agent follows to try to achieve its goals and handle events•event: occurrence to which agent should respond


JACK – Agent Templateagent AgentType extends Agent {

// Knowledge bases used by the agent are declared here.

#private data BeliefType belief_name(arg_list);

// Events handled, posted and sent by the agent are declared here.

#handles event EventType;

#posts event EventType reference; used to create internal events

#sends event EventType reference; used to send messages to other agents

// Plans used by the agent are declared here. Order is important.

#uses plan PlanType;

// Capabilities that the agent has are declared here.

#has capability CapabilityType reference;

// other Data Member and Method definitions

}


Jason – Features• beliefs: weak and strong negation to support both closed-world

assumption and open-world

• belief annotations: label information source, e.g. self, percept

• events: internal, messages, percepts

• a library of “internal actions”, e.g. send

• user-defined internal actions: programmed in Java.

• automatic handling of plan failures

• annotations on plan labels: used to select a plan

• speech-act based inter-agent communication

• Java-based customization: (plan) selection functions, trust functions, perception, belief-revision, agent communication


Jason – Planstriggering event

test on beliefs

plan body


Summary

Key language elements of APLs:

• beliefs and goals to represent environment

• events received from environment (& internal)

• actions to update beliefs, adopt goals, send messages, act in environment

• plans, capabilities & modules to structure action

• rules to select actions/plans/modules/capabilities

• support for multi-agent systems


How are these APLs related?

AGENT-01

(PLACA )

Family of LanguagesBasic concepts: beliefs, action, plans, goals-to-do):

AgentSpeak(L)1, Jason2

Golog 3APL3

＝

＝

＝

1 mainly interesting from a historical point of view2 from a conceptual point of view, we identify AgentSpeak(L) and Jason3 without practical reasoning rules

Main addition: Declarative goals

2APL ≈ 3APL + GOAL

A comparison from a high-level, conceptual point, not taking into account any practical aspects (IDE, available docs, speed, applications, etc)

Java-based BDI Languages

Agent Factory, Jack (commercial), Jadex, JIAC

Mobile Agents

CLAIM, AgentScape

Multi-Agent SystemsAll of these languages

(except AGENT-0, PLACA, JACK) have versions implemented

“on top of” JADE.

Pro

log

-bas

ed


ReferencesWebsites• 2APL: http://www.cs.uu.nl/2apl/ • Agent Factory: http://www.agentfactory.com • GOAL: http://mmi.tudelft.nl/trac/goal• JACK: http://www.agent-software.com.au/products/jack/• Jadex: http://jadex.informatik.uni-hamburg.de/• Jason: http://jason.sourceforge.net/• JIAC: http://www.jiac.de/

Books• Bordini, R.H.; Dastani, M.; Dix, J.; El Fallah Seghrouchni, A. (Eds.), 2005 Multi-

Agent Programming Languages, Platforms and Applications. presents 3APL, CLAIM, Jadex, Jason

• Bordini, R.H.; Dastani, M.; Dix, J.; El Fallah Seghrouchni, A. (Eds.), 2009, Multi-Agent Programming: Languages, Tools and Applications.presents a.o.: Brahms, CArtAgO, GOAL, JIAC Agent Platform

Koen Hindriks Multi-Agent Systems

The GOALAgent Programming Language

Koen Hindriks Multi-Agent Systems 2012 20 Koen Hindriks Multi-Agent Systems

THE BLOCKS WORLDThe Hello World example of Agent Programming


The Blocks World

• Positioning of blocks on table is not relevant.• A block can be moved only if it there is no other block on top of it.

Objective: Move blocks in initial state such that result is goal state.

A classic AI planning problem.


Representing the Blocks World

Basic predicate:• on(X,Y).

Defined predicates:• tower([X]) :- on(X,table).

tower([X,Y|T) :- on(X,Y),tower([Y|T]).• clear(X) :- block(X), not(on(Y,X)).• clear(table).• block(X) :- on(X, _).

EXERCISE:

Prolog is the knowledge representation language used in GOAL.


Representing the Initial State

Using the on(X,Y) predicate we can represent the initial state.

beliefs{ on(a,b), on(b,c), on(c,table), on(d,e), on(e,table), on(f,g), on(g,table).}

Initial belief base of agent


Representing the Blocks World• What about the rules we defined before?• Add clauses that do not change into the knowledge base.

tower([X]) :- on(X,table).tower([X,Y|T]) :- on(X,Y),tower([Y|T]).clear(X) :- block(X), not(on(Y,X)). clear(table).block(X) :- on(X, _).

knowledge{ block(X) :- on(X, _). clear(X) :- block(X), not(on(Y,X)). clear(table). tower([X]) :- on(X,table). tower([X,Y|T]) :- on(X,Y), tower([Y|T]).}

Static knowledge base of agent


Why a Separate Knowledge Base?

• Concepts defined in knowledge base can be used in combination with both the belief and goal base.

• Example– Since agent believes on(e,table),on(d,e), then infer:

agent believes tower([d,e]).– If agent wants on(a,table),on(b,a), then infer: agent

wants tower([b,a]).

• Knowledge base introduced to avoid duplicating clauses in belief and goal base.


Representing the Goal State

Using the on(X,Y) predicate we can represent the goal state.

goals{ on(a,e), on(b,table), on(c,table), on(d,c), on(e,b), on(f,d), on(g,table).}

Initial goal base of agent


One or Many Goals

In the goal base using the comma- or period-separator makes a difference!

goals{ on(a,table), on(b,a), on(c,b).}

goals{ on(a,table). on(b,a). on(c,b).}

• Left goal base has three goals, right goal base has single goal.

• Moving c on top of b (3rd goal), c to the table, a to the table (2nd goal) , and b on top of a (1st goal) achieves all three goals but not single goal of right goal base.

• The reason is that the goal base on the left does not require block c to be on b, b to be on a, and a to be on the table at the same time.


Mental State of GOAL Agent

knowledge{ block(X) :- on(X, _). clear(X) :- block(X), not(on(Y,X)). clear(table). tower([X]) :- on(X,table). tower([X,Y|T]) :- on(X,Y), tower([Y|T]).}beliefs{ on(a,b), on(b,c), on(c,table), on(d,e), on(e,table), on(f,g), on(g,table).}goals{ on(a,e), on(b,table), on(c,table), on(d,c), on(e,b), on(f,d), on(g,table).}

The knowledge, belief, and goal sections together constitute the specification of the Mental State of a GOAL Agent.

Initial mental state of agent


Inspecting the Belief & Goal base

• Operator bel()to inspect the belief base.

• Operator goal()to inspect the goal base.– Where is a Prolog conjunction of literals.

• Examples:– bel(clear(a), not(on(a,c))).– goal(tower([a,b])).


Inspecting the Belief Base• bel() succeeds if follows from the belief base

in combination with the knowledge base.

• Example:– bel(clear(a), not(on(a,c))) succeeds

• Condition is evaluated as a Prolog query.

knowledge{ block(X) :- on(X, _). clear(X) :- block(X), not(on(Y,X)). clear(table). tower([X]) :- on(X,table). tower([X,Y|T]) :- on(X,Y), tower([Y|T]).}beliefs{ on(a,b), on(b,c), on(c,table), on(d,e), on(e,table), on(f,g), on(g,table).}


Inspecting the Belief Base

Which of the following succeed?1.bel(on(b,c), not(on(a,c))).

2.bel(on(X,table), on(Y,X), not(clear(Y)).

3.bel(tower([X,b,d]).

[X=c;Y=b]


EXERCISE:


Inspecting the Goal Base

• goal() succeeds if follows from one of the goals in the goal base in combination with the knowledge base.

• Example:– goal(clear(a))succeeds.– but not goal(clear(a),clear(c)).

Use the goal(…) operator to inspect the goal base.

knowledge{ block(X) :- on(X, _). clear(X) :- block(X), not(on(Y,X)). clear(table). tower([X]) :- on(X,table). tower([X,Y|T]) :- on(X,Y), tower([Y|T]).}goals{ on(a,e), on(b,table), on(c,table), on(d,c), on(e,b), on(f,d), on(g,table).}


Inspecting the Goal Base

Which of the following succeed?1.goal(on(b,table), not(on(d,c))).

2.goal(on(X,table), on(Y,X), clear(Y)).

3.goal(tower([d,X]).

knowledge{ block(X) :- on(X, _). clear(X) :- block(X), not(on(Y,X)). clear(table). tower([X]) :- on(X,table). tower([X,Y|T]) :- on(X,Y), tower([Y|T]).}goals{ on(a,e), on(b,table), on(c,table), on(d,c), on(e,b), on(f,d), on(g,table).}

EXERCISE:


Negation and Beliefs

not(bel(on(a,c))) = bel(not(on(a,c)))?

• Answer: Yes.– Because Prolog implements negation as failure.

– If φ cannot be derived, then not(φ) can be derived.

– We always have: not(bel()) = bel(not())


EXERCISE:


Negation and Goals

not(goal()) = goal(not())?

• Answer: No.

• We have, for example:goal(on(a,b)) and goal(not(on(a,b))).

knowledge{ block(X) :- on(X, _). clear(X) :- block(X), not(on(Y,X)). clear(table). tower([X]) :- on(X,table). tower([X,Y|T]) :- on(X,Y), tower([Y|T]).}goals{ on(a,b), on(b,table). on(a,c), on(c,table).}

EXERCISE:


Combining Beliefs and Goals

• Achievement goals:– a-goal() = goal(), not(bel())

• Agent only has an achievement goal if it does not believe the goal has been reached already.

• Goal achieved:– goal-a() = goal(), bel()

• A (sub)-goal has been achieved if the agent believes in .

Useful to combine the bel(…) and goal(…) operators.


Expressing BW Concepts

• Define: block X is misplaced• Solution:

goal(tower([X|T])),not(bel(tower([X|T]))).

• But this means that saying that a block is misplaced is saying that you have an achievement goal:

a-goal(tower([X|T])).

Possible to express key Blocks World concepts by means of basic operators.

Mental States

EXERCISE:


ACTIONS SPECIFICATIONSChanging Blocks World Configurations


Actions Change the Environment…

move(a,d)


and Require Updating Mental States.• To ensure adequate beliefs after performing an action the belief base

needs to be updated (and possibly the goal base).

– Add effects to belief base: insert on(a,d) after move(a,d).– Delete old beliefs: delete on(a,b) after move(a,d).


and Require Updating Mental States.• If a goal has been (believed to be) completely achieved, the goal is

removed from the goal base.

• It is not rational to have a goal you believe to be achieved.• Default update implements a blind commitment strategy.

move(a,b)

beliefs{ on(a,table), on(b,table).}goals{ on(a,b), on(b,table).}

beliefs{ on(a,b), on(b,table).}goals{ }


Action Specifications• Actions in GOAL have preconditions and

postconditions.• Executing an action in GOAL means:

– Preconditions are conditions that need to be true:• Check preconditions on the belief base.

– Postconditions (effects) are add/delete lists (STRIPS):• Add positive literals in the postcondition• Delete negative literals in the postcondition

• STRIPS-style specificationmove(X,Y){ pre { clear(X), clear(Y), on(X,Z), not( on(X,Y) ) } post { not(on(X,Z)), on(X,Y) }}


move(X,Y){

pre { clear(X), clear(Y), on(X,Z), not( on(X,Y) )}

post { not(on(X,Z)), on(X,Y) }

}

Example: move(a,b)• Check: clear(a), clear(b), on(a,Z), not( on(a,b) )• Remove: on(a,Z)• Add: on(a,b)

Note: first remove, then add.

Actions Specifications

table


move(X,Y){

pre { clear(X), clear(Y), on(X,Z) }


}

Example: move(a,b)

Actions Specifications

beliefs{ on(a,table), on(b,table).}

beliefs{ on(b,table). on(a,b).}


move(X,Y){

pre { clear(X), clear(Y), on(X,Z) }


}

1. Is it possible to perform move(a,b)?

2. Is it possible to perform move(a,d)?

Actions SpecificationsEXERCISE:

knowledge{ block(a), block(b), block(c), block(d), block(e), block(f), block(g), block(h), block(i). clear(X) :- block(X), not(on(Y,X)). clear(table). tower([X]) :- on(X,table). tower([X,Y|T]) :- on(X,Y), tower([Y|T]).}beliefs{ on(a,b), on(b,c), on(c,table), on(d,e), on(e,table), on(f,g), on(g,table).}

No, not( on(a,b) ) fails. Yes.


ACTION RULESSelecting actions to perform


Agent-Oriented Programming

• How do humans choose and/or explain actions?

• Examples:• I believe it rains; so, I will take an umbrella with me.• I go to the video store because I want to rent I-robot.• I don’t believe busses run today so I take the train.

• Use intuitive common sense concepts:

beliefs + goals => action

See Chapter 1 of the Programming Guide


Selecting Actions: Action Rules

• Action rules are used to define a strategy for action selection.

• Defining a strategy for blocks world:– If constructive move can be made, make it.– If block is misplaced, move it to table.

• What happens:– Check condition, e.g. can a-goal(tower([X|T]))be derived given

current mental state of agent?– Yes, then (potentially) select move(X,table).

program{ if bel(tower([Y|T])), a-goal(tower([X,Y|T])) then move(X,Y). if a-goal(tower([X|T])) then move(X,table).}


Order of Action Rules

• Action rules are executed by default in linear order.• The first rule that fires is executed.

• Default order can be changed to random.• Arbitrary rule that is able to fire may be selected.

program{ if bel(tower([Y|T])), a-goal(tower([X,Y|T])) then move(X,Y). if a-goal(tower([X|T])) then move(X,table).}

program[order=random]{ if bel(tower([Y|T])), a-goal(tower([X,Y|T])) then move(X,Y). if a-goal(tower([X|T])) then move(X,table).}


Example Program: Action RulesAgent program may allow for multiple action choices

dTo table

Random, arbitrary choice

program[order=random]{ if bel(tower([Y|T])), a-goal(tower([X,Y|T])) then move(X,Y). if a-goal(tower([X|T])) then move(X,table).}


The Sussman Anomaly (1/5)

• Non-interleaved planners typically separate the main goal, on(A,B),on(B,C) into 2 sub-goals: on(A,B) and on(B,C).

• Planning for these two sub-goals separately and combining the plans found does not work in this case, however.

a

c

Initial state

b c

b

a

Goal state


The Sussman Anomaly (2/5)• Initially, all blocks are misplaced• One constructive move can be made (c to table)• Note: move(b,c) not enabled.• Only action enabled: c to table (2x).

Need to check conditions of action rules: if bel(tower([Y|T]),a-goal(tower([X,Y|T))then move(X,Y). if a-goal(tower([X|T))then move(X,table).

We have bel(tower([c,a]) and a-goal(tower([c])).

c

b

a

Goal state

a

c

Initial state

b


The Sussman Anomaly (3/5)• Only constructive move enabled is

– Move b onto c

Need to check conditions of action rules: if bel(tower([Y|T]), a-goal(tower([X,Y|T))then move(X,Y). if a-goal(tower([X|T))then move(X,table).

Note that we have:

a-goal(on(a,b),on(b,c),on(c,table)),but not: a-goal(tower[c])).

Current state

c

b

a

Goal state

ac b


The Sussman Anomaly (4/5)• Again, only constructive move enabled

– Move a onto b

Need to check conditions of action rules: if bel(tower([Y|T]), a-goal(tower([X,Y|T))then move(X,Y). if a-goal(tower([X,T))then move(X,Y).

Note that we have: a-goal(on(a,b),on(b,c),on(c,table)), but not: a-goal(tower[b,c]).

c

b

a

Goal state

ac

b

Current state


The Sussman Anomaly (5/5)• Upon achieving a goal completely

that goal is automatically removed.• The idea is that no resources should

be wasted on achieving the goal.

In our case, goal(on(a,b),on(b,c),on(c,table)) has been

achieved, and is dropped. The agent has no other

goals and is ready.

c

b

a

Goal state

a

c

b

Current state


Organisation• Read Programming Guide Ch1-3 (+ User Manual)

• Tutorial:– Download GOAL: See http://ii.tudelft.nl/trac/goal (v4537)– Practice exercises from Programming Guide– BW4T assignments 3 and 4 available

• Next lecture:– Sensing, perception, environments– Other types of rules & macros– Agent architectures

introduction agent programming

Documents