BDI Agents: From Theory to Practice
Anand S. Rao and Michael P. Georgeff
Australian Artificial Intelligence Institute
Level 6, 171 La Trobe Street
Melbourne, Australia
Email" anand@aaii.oz.au and georget~aaii.oz.au
Abstract
The study of computational agents capable of
rational behaviour has received a great deal of
attention in recent years. Theoretical formal-
izations of such agents and their implementa-
tions have proceeded in parallel with little or
no connection between them. Tkis paper ex-
plores a particular type of rational agent, a Belief-
Desire-Intention (BDI) agent. The primary aim
of this paper is to integrate (a) the theoretical
foundations of BDI agents from both a quantita-
tive decision-theoretic perspective and a symbolic
reasoning perspective; (b) the implementations
of BDI agents from an ideal theoretical perspec-
tive and a more practical perspective; and (c) the
building of large-scale applications based on BDI
agents. In particular, an air-trafflc management
application will be described from both a theo-
retical and an implementation perspective.
Introduction
The design of systems that are required to perform
high-level management and control tasks in complex
dynamic environments is becoming of increasing com-
mercial importance. Such systems include the man-
agement and control of air traffic systems, telecommu-
nications networks, business processes, space vehicles,
and medical services. Experience in applying conven-
tional software techniques to develop such systems has
shown that they are very difficult and very expensive
to build, verify, and maintain. Agent-oriented systems,
based on a radically different view of computational
entities, offer prospects for a qualitative change in this
position.
A number of different approaches have emerged
as candidates for tile study of agent-oriented sys-
tems (Bratman et al. 1988; Doyle 1992; Rao and
Georgeff 1991c; Rosenschein and Kaelbling 1986;
Shoham 1993). One such architecture views the sys-
tem as a rational agent having certain mental attttudes
of Belief, Desire and Intention (BDI), representing, re-
spectively, the information, motivational, and deliber-
ative states of the agent. These mental attitudes deter-
mine the system’s behaviour and are critical for achiev-
ing adequate or optimal performance when delibera-
tion is subject to resource bounds (Bratman 1987;
Kinny and Georgeff 1991).
While much work has gone into the formaliza-
tion (Cohen and Levesque 1990; Jennings 1992; Kinny
et al. 1994; Rao and Georgeff 1991c; Singh and Asher
1990) and implementation (Burmeister and Sunder-
meyer 1992; Georgeff and Lansky 1986; Muller et al.
1994; Shoham 1993) of BDI agents, two main criticisms
have been levelled against these endeavours. First,
the having of these three attitudes is attacked from
both directions: classical decision theorists and plan-
ning researchers question the necessity of having all
three attitudes and researchers from sociology and Dis-
tributed Artificial Intelligence question the adequacy
of these three alone. Second, the utility of studying
multi-modal BDI logics which do not have complete
axiomatizations and are not efficiently computable is
questioned by many system’ builders as having little
relevance in practice.
This paper addresses these two criticisms from the
perspectives of the authors’ previous work in BDI
logics (Rao and Georgeff 1991a; 1991c; 1993), sys-
tems (Georgeff and Lansky 1986), and real-world ap-
plications (Ingrand et al. 1992; Rao et al. 1992).
We argue the necessity (though not the adequacy)
these three attitudes in domains where real-time per-
formance is required from both a quantitative decision-
theoretic perspective and a symbolic reasoning per-
spective. To address the second criticism, we show
how one can build practical systems by making certain
simplifying assumptions and sacrificing some of the ex-
pressive power of the theoretical framework. We first
describe a practical BDI interpreter and show how it
relates to our theoretical framework. We then describe
an implemented agent-oriented air-traffic managemeut
system, called OASIS, currently being tested at Syd-
ney airport.
The primary purpose of this paper is to provide a
unifying framework for a particular type of agent, BDI
agent, by bringing together various elements ofour pre-
vious work in theory, systems, and applications.
312 ICMAS-9S
From: Proceedings of the First International Conference on Multiagent Systems. Copyright © 1995, AAAI (www.aaai.org). All rights reserved.

The System and its Environment
We first informally establish the necessity of beliefs,
desires, and intentions for a system to act appropri-
ately in a class of application domains characterized
by various practical limitations and requirements. As
typical of such a domain, consider the design of an air
traffic management system that is to be responsible
for calculating the expected time of arrival (ETA) for
arriving aircraft, sequencing them according to certain
optimality criteria, reassigning the ETA for the aircraft
according to the optimal sequence, issuing control di-
rectives to the pilots to achieve the assigned ETAs, and
monitoring conformance.
This and a wide class of other real-time application
domains exhibit a number of important characteristics:
1. At any instant of time, there are potentially many
different ways in which the environment can evolve
(formally, the environment is nondeterministic); e.g.,
the wind field can change over time in unpredictable
ways, as can other parameters such as operating
conditions, runway conditions, presence of other air-
craft, and so on.
2. At any instant of time, there are potentially many
different actions or procedures the system can exe-
cute (formally, the system itself is nondeterministic);
e.g., the system can take a number of different ac-
tions, such as requesting an aircraft change speed,
stretch a flight path, shorten a flight path, hold, and
80 on.
3. At any instant of time, there are potentially many
different objectives that the system is asked to ac-
complish; e.g., the system can be asked to land air-
craft QF001 at time 19:00, land QF003 at 19:01, and
maximize runway throughput, not all of which may
be simultaneously achievable.
4. The actions or procedures that (best) achieve the
various objectives are dependent on the state of the
environment (context) and are independent of the
internal state of the system; e.g., the actions by
which the aircraft achieve their prescribed landing
times depend on wind field, operating conditions,
other aircraft, and so on, but not on the state of the
computational system.
5. The environmeiit can only be sensed locally (i.e., one
.sensing action is not sufficient for fully determining
the state of the entire environment); e.g., the system
receives only spot wind data from some aircraft at
some times at some locations and thus cannot de-
termine in one sensing operation the current wind
field.
6. The rate at which computations and actions can
be carried out is within reasonable bounds to the
rate at which the environment evolves; e.g., changes
in wind field, operational conditions, runway con-
ditions, presence of other aircraft, and so on, can
occur during the calculation of an efficient landing
sequence and during the period that the aircraft is
flying to meet its assigned landing time.
One way of modelling the behaviour of such a sys-
tem, given Characteristics (1) and (2), is as a branch-
ing tree structure (Emerson 1990), where each branch
in the tree represents a’n alternative execution path.
Each node in the structure represents a certain state
of the world, and each transition a primitive action
made by the system, a primitive event occurring in the
environment, or both.
If we differentiate the actions taken by the system
and the events taking place in the environment, the two
different types of nondeterminism manifest themselves
in two different node types. We call these choice (deci-
sion) nodes and chance nodes, representing the options
available to the system itself and the uncertainty of the
environment, respectively.
In this formal model, we can identify the objectives
of the system with particular paths through the tree
structure, each labeled with the objective it realizes
and, if necessary, the benefit or payoff obtained by
traversing this path.
As the system has to act, it needs to select appro-
priate actions or procedures to execute from the var-
ious options available to it. The design of such a se-
lection function should enable the system to achieve
effectively its primary objectives, given the computa-
tional resources available to the system and the char-
acteristics of the environment in which the system is
situated.
Under the above-mentioned domain characteristics,
there are at least two types of input data required by
such a selection function. First, given Characteristic
(4), it is essential that the system have information
on the state of the environment~ But as this cannot
necessarily be determined in one sensing action (Char-
acteristics 1 and 5), it is necessary that there be some
component of the system that can represent this infor-
mation and is updated appropriately after each sensing
action. We call such a component the system’s beliefs.
This component may be implemented as a variable, a
database, a set of logical expressions, or some other
data structure. Thus, beliefs can be viewed as the in-
formative component of system state.1
Second, it is necessary that the system also have in-
formation about the objectives to be accomplished or,
more generally, what priorities or payoffs are associated
with the various current objectives (Characteristics
and 4). It is possible to think of these objectives, or
their priorities, as being generated instantaneously or
1We distinguish beliefs from the notion of knowledge,
as defined for example in the literature on distributed com-
puting, as the system beliefs are only required to provide
information on the likely state of the environment; e.g.,
certain assumptions may be implicit in the implementation
but sometimes violated in practice, such as assumptions
about accuracy of sensors, or rate of change of certain en-
vironmental conditions.
Rao 313
From: Proceedings of the First International Conference on Multiagent Systems. Copyright © 1995, AAAI (www.aaai.org). All rights reserved.

functionally, and thus not requiring any state repre-
sentation (unlike the system beliefs, which cannot be
represented functionally). We call this component the
system’s desires, which can be thought of as represent-
ing the motivational state of the system .2
Given this picture, the most developed approach rel-
evant to the design of the selection function is decision
theory. However, the decision-theoretic approach does
not take into account Characteristic (6); namely, that
the environment may change in possibly significant and
unanticipated ways either (a) during execution of the
selection function itself or (b) during the execution
the course of action determined by the selection func-
tion.
The possibility of the first situation arising can be
reduced by using a faster (and thus perhaps less opti-
mal) selection functio’n, as there is then less risk of
significant event occurring during computation.
Interestingly, to the second possibility, classical de-
cision theory and classical computer science provide
quite different answers: decision theory demands that
one re-apply the selection function in the changed envi-
ronment; standard computer programs, once initiated,
expect to execute to completion without any reassess-
ment of their utility.
Given Characteristic (6), neither approach is satis-
factory, l~-application of the selection function in-
creases substantially the risk that significant changes
will occur during this calculation and also consumes
time that may be better spent in action towards achiev-
ing the given objectives. On the other hand, execution
of any course of action to completion increases the risk
that a significant change will occur during this execu-
tion, the system thus failing to achieve the intended
objective or realizing the expected utility.
We seem caught on the horns of a dilemma: re-
considering the choice of action at each step is po-
tentially too expensive and the chosen action pos-
sibly invalid, whereas unconditional commitment to
the chosen course of action can result in the system
failing to achieve its objectives. However, assum-
ing that potentially significant changes can be deter-
mined instantaneously, s it is possible to limit the fre-
quency of reconsideration and thus achieve an appro-
priate balance between too much reconsideration and
not enough (Kinny and Georgeff 1991). For this
work, it is necessary to include a component of system
state to represent the currently chosen course of action;
that is, the output of the most recent call to the selec-
tion function. We call this additional state component
the system’s intentions. In essence, the intentions of
the system capture the deliberative component of the
system.
2We distinguish desires from goals as they are defined,
for example, in the AI literature in that they may be many
at any instant of time and may be mutually incompatible.
3"rhat is, at the level of granularity defined by tile prim-
itive actions and events of the domain.
314 ICMAS-95
Decision Trees to Possible Worlds
While in the previous section we talked abstractly
about the belief, desire, and intention components of
the system state, in this section we develop a theory for
describing those components in a propositional form.
We begin with classical decision theory and show how
we can view such a theory within a framework that
is closer to traditional epistemic models of belief and
agency. In later sections, we will show how this model
can then be used to specify and implement systems
with the characteristics described above.
Informally, a decision tree consists of decision nodes,
chance nodes, and terminal nodes, and includes a prob-
ability function that maps chance nodes to real-valued
probabilities (including conditional probabilities) and
a payoff function that maps terminal nodes to real
numbers. A deliberation function, such as maximin or
maximizing expected utility is then defined for choosing
one or more best sequences of actions to perform at a
given node.
We transform such a decision tree, and appropri-
ate deliberation functions, to an equivalent model that
represents beliefs, desires, and intentions as separate
accessibility relations over sets of possible worlds. This
transformation provides an alternative basis for cases
in which we have insufficient information on probabil-
ities and payoffs and, perhaps more importantly, for
handling the dynamic aspects of the problem domain.
We begin by considering a fulldecision tree, in which
every possible path is represented (including those with
zero payoffs). Given such a decision tree, we start from
the root node and traverse each arc. For each unique
state labeled on an arc emanating from a chance node,
we create a new decision tree that is identical to the
original tree except that (a) t]~e chance node is re-
moved and (b) the arc incident on the chance node
connected to the successor of the chance node. This
process is carried out recursively until there are no
chance nodes left. This yields a set of decision trees,
each consisting of only decision nodes and terminal
nodes, and each corresponding to a different possi-
ble state of the environment. That is, from a tra-
ditional possible-worlds perspective, each of these de-
cision trees represents a different possible world with
different probability of occurrence. Finally, the payoff
function is assigned to paths in a straightforward way.
The algorithm for this transformation can be found
elsewhere (B.ao and Georgeff 1991b).
The resulting possible-worlds model contains two
types of information, represented by the probabili-
ties across worlds and the payoffs assigned to paths.
We now split these out into two accessibility rela-
tions, the probabilities being represented in the belief-
accessibility relation and the payoffs in the desire-
accessibility relation. The sets of tree structures de-
fined by these relations are identical, although with-
out loss of generality we could delete from the desire-
accessible worlds all paths with zero payoffs.
From: Proceedings of the First International Conference on Multiagent Systems. Copyright © 1995, AAAI (www.aaai.org). All rights reserved.

Given a decision tree and the above transformation,
an agent can now make use of the chosen deliberation
function to decide the best course(s) of action. We can
formally represent these selected path(s) in the deci-
sion tree using a third accessibility relation on possible
worlds, corresponding to the intentions of the agent.
In essence, for each desire-accessible world, there exists
a corresponding intention-accessible world which con-
tains only the best course(s) of action as determined
by the appropriate deliberation function.
Thus, our possible-worlds model consists of a set
of possible worlds where each possible world is a
tree structure. A particular index within a possi-
ble world is called a situation. With each situation
we associate a set of belie]-accessible worlds, desire-
accessible worlds, and intention.accessible worlds; in-
tuitively, those worlds that the agent believes to be
possible, desires to bring about; and intends to bring
about, respectively.
BDI Logics
The above transformation provides the basis for devel-
oping a logical theory for deliberation by agents that is
compatible with quantitative decision theory in those
cases where we have good estimates for probabilities
and payoffs. However, it does not address the case in
which we do not have such estimates, nor does it ad-
dress the dynamic aspects of deliberation, particularly
those concerning commitment to previous decisions.
We begin by abstracting the model given above to
reduce probabilities and payoffs to dichotomous (0-1)
values. That is, we consider propositions to be either
believed or not believed, desired or not desired, and
intended or not intended, rather than ascribing con-
tinuous measures to them. Within such a framework,
we first look at the static properties we would want of
BDI systems and next their dynamic properties.
The axiomatization for beliefs that we adopt is the
standard weak-S5 (or KD45) modal system. We adopt
the D and K axioms for desires and intentions; i.e., de-
sires and intentions have to be closed under implication
and have to be consistent. We also have the inference
rule of necessitation for beliefs, desires, and intentions.
A number of researchers have proposed their pre-
ferred axiomatizations capturing the relationships be-
tween beliefs, desires, and intentions. However, in
other work (Rao and Georgeff 1991c) we depart from
this approach and give a comprehensive family of BDI
logics similar in tradition to that of modal logic sys-
tems (i.e., KD45 system, $4 system, etc.). The reason
for this departure is that we do not believe that there
need be a unique and correct axiomatization that cov-
ers all interesting BDI agents---one may want to model
different types df agents for different purposes.
Static Constraints: The static relationships
among the belief-, desire-, and intention-accessible
worlds can be examined along two different dimen-
sions, one with respect to the sets of possible worlds
and the other with respect to the structure of the possi-
ble worlds. Given two relations four different relation-
ships are possible between them: one being a subset of
the other and vice versa, and their intersections being
null or non-nulL Similarly, as each possible world is a
time tree, there are four possible structural relation-
ships that can hold between any pair of worlds: one
could be a sub-world of the other or vice versa, or the
worlds could be identical or incomparable.
Now we can combine the set and structural rela-
tionships of the belief, desire, and intention worlds to
obtain twelve different BDI systems. Some of these
relationships and axiomatizations can be derived from
the others. Three of the above relationships and ax-
iomatizations have been considered before under the
terms realism (Cohen and Levesque 1990) (if an agent
believes a proposition, it will desire it), strong real-
ism (Rao and Georgeff 1991c) (if an agent desires
achieve a proposition, it will believe the proposition
to be an option) and weak realism (Rao and Georgeff
1991a) (if an agent desires to achieve a proposition,
will not believe the negation of the proposition to be
inevitable).
The choice of BDI system depends also on which
other properties are desired of the agent. For example,
a number of researchers have proposed requirements
concerning the asymmetry between beliefs and other
attitudes (Bratman 1987; Rao and Georgeff 1991a) and
consequential closure principles (Cohen and Levesque
1990). The first requires that rational agents maintain
consistency between their beliefs, desires, and inten-
tions, but not completeness. The second requires that
the beliefs, desires, and intentions of an agent must
not be closed under the implications of the other at-
titudes. All the above properties are satisfied by a
BDI system in which the pair-wi~e intersections of the
belief-, desire-, and intention-accessible worlds are non-
null. Other BDI systems in which intention-accessible
worlds axe sub-worlds of desire-accessible worlds, which
are sub-worlds of belief-accessible worlds satisfy some,
but not all of these properties.
Dynamic Constraints: As discussed earlier, an
important aspect of a BDI architecture is the notion
of commitment to previous decisions. A commitment
embodies the balance between the reactivity and goal-
directedness of an agent-oriented system. In a con-
tinuously changing environment, commitment lends a
certain sense of stability to the reasoning process of an
agent. This results in savings in computational effort
and hence better overall performance (Bratman 1987;
Kinny and Georgeff 1991; Bao and Georgeff 1991c).
A commitment usually has two parts to it: one is
the condition that the agent is committed to maintain,
called the commitment condition, and the second is the
condition under which the agent gives up the commit-
ment, called the termination condition. As the agent
has no direct control over its beliefs and desires, there
is no way that it can adopt or effectively realize a corn-
Rao 31S
From: Proceedings of the First International Conference on Multiagent Systems. Copyright © 1995, AAAI (www.aaai.org). All rights reserved.

mitment strategy over these attitudes. However, an
agent can choose what to do with its intentions. Thus,
we restrict the commitment condition to intentions.
An agent can commit to an intention based on the ob-
ject of the intention being fulfilled in one future path or
all future paths leading to different commitment con-
ditions and hence different dynamic behaviours.
Different termination conditions result in further
variations in behaviour(Rao and Georgeff 1991c; 1993;
Georgeff and Rao August 1995). For example, we
can define a blindly-committed agent which denies any
changes to its beliefs or desires that would conflict with
its commitments; a single-minded agent which enter-
talns changes to beliefs and will drop its commitments
accordingly; and an open-minded agent which allows
changes in both its beliefs and desires that will force
its commitments to be dropped.
The various forms of termination and commitment
can be expressed as axioms of our logic, and semantic
constraints can be placed on the dynamic evolution
of the accessibility relations. As before, rather than
claiming that one particular commitment strategy is
the right strategy, we allow the user to tailor them
according to the application.
The purpose of the a~ove formalization is to build
formally verifiable and pr,’~tical systems. If for a given
application domain, we 1~ ~ow how the environment
changes and the behaviour~ expected of the system, we
can use such a formalization to specify, design, and ver-
ify agents that, when placed in such an environment,
will exhibit all and only the desired behaviours. Else-
where (Rao and Georgeff 1993) we have described how
to verify certain behaviours of agents based on their
static constraints and their commitment strategies us-
ing a model-checking approach. In the next section, we
turn ~o the task of building a practical system based
on the above theory.
Abstract Architecture
While it is not necessary that a system that is spec-
ified in terms of beliefs, desires and intentions bede-
signed with identifiable data structures corresl~onding
to each of these components, the architecture we pro-
pose below is based on such a correspondence. The
rationale for such a design is that the identification of
beliefs, desires, and intentions is useful when the sys-
tem must communicate with humans or other software
agents and can be expected to simplify the building,
maintenance, and verification of application systems.
On the other hand, the architecture cannot be sim-
ply based on a traditional theorem-proving system,
even if extended to handle the temporal, epistemic,
and non-deterministic elements of the logic described
above. The reason for this is that the time taken to
reason in this way, and thus the time taken to act,
is potentially unbounded, thereby destroying the reac-
tivity that is essential to an agent’s urvival. Thus,
although we could use a theorem prover to reason "off-
316 ICMAS-95
line" about the behaviour of an agent-based system,
we cannot directly use such a theorem prover to im-
plement the system itself.
The abstract architecture we propose comprises
three dynamic data structures representing the agent’s
beliefs, desires, and intentions, together with an input
queue of events. We allow update and query operations
on the three data structures. The update operations
on beliefs, desires, and intentions are subject to re-
spective compatibility requirements. These functions
are critical in enforcing the formalized constraints upon
the agent’s mental attitudes as described before. The
events the system can recognize include both external
events and internal events. We assume that the events
are atomic and are recognized after they have occurred.
Similarly, the outputs of the agent--actions--are also
assumed to be atomic. The main interpreter loop is
given below. We assume that the event queue, belief,
desire, and intention structures are global.
BDI-interpreter
initialize-state();
repeat
options := option-generator(event-queue);
selected-options := deliberate(options);
update-intentions(selected-options);
execute();
get-new-external-events();
drop-successful-attitudes0;
drop-impossible-attitudes();
end repeat
At the beginning of every cycle, the option genera-
tor reads the event queue and returns a list of options.
Next, the deliberator selects a subset of options to be
adopted and adds these to the intention structure. If
there is an intention to perform a~n atomic action at this
point in time, the agent then executes it. Any external
events that have occurred during the interpreter cycle
are then added to the event queue. Internal events are
added as they occur. Next, the agent modifies the in-
tention and desire structures by dropping all successful
desires and satisfied intentions, as well as impossible
desires and unrealizable intentions.
This abstract architecture is an idealization that
faithfully captures the theory, including the various
components of practical reasoning (Bratman 1987);
namely, option generation, deliberation, execution,
and intention handling. However, it is not a prac-
tical system for rational reasoning. The architecture
is basod on a (logically) closed set of beliefs, desires,
and intentions and the provability procedures required
arc not computable. Moreover, wc have given no in-
dication of how the option generator and deliberation
procedures can be made sufficiently fast to satisfy the
real-time demands placed upon the system.
We therefore make a number of important choices
of representation which, while constraining expressive
power, provide a more practical system for rational rea-
soning. The system presented is a simplified version of
From: Proceedings of the First International Conference on Multiagent Systems. Copyright © 1995, AAAI (www.aaai.org). All rights reserved.

the Procedural Reasoning System (PRS) (Georgeff
Lansky 1986; Ingrand et a/. 1992), one of the first im-
plemented agent-oriented systems based on the BDI
architecture, and a successor system, dMARS (dis-
tributed MuitiAgent Reasoning System).
First, we explicitly represent only beliefs about the
current state of the world and consider only ground sets
of literais with n’o disjunctions or implications. Intu-
itively, fhese represent beliefs that are currently held,
but which can be expected to change over time.
Second, we represent the information about the
means of achieving certain future world states and the
options available to the agent as plans, which can be
viewed as a special form of beliefs (Rao and Georgeff
1992). Intuitively, plans are abstract specifications of
both the means for achieving certain desires and the
options available to the agent. Each plan has a body
describing the primitive actions or subgoals that have
to be achieved for plan execution to be successful. The
conditions under which a plan can be chosen as an op-
tion are specified by an invocation condition and a pre-
condition; the invocation condition specifies the "trig-
gering" event that is necessary for invocation of the
plan, and the precondition specifies the situation that
must hold for the plan to be executable.
Third, each intention that the system forms by
adopting certain plans of action is represented implic-
itly using a conventional run-time stack of hierarchi-
cally related plans (similar to how a Prolog interpreter
handles clauses). 4 Multiple intention stacks can co-
exist, either running in parallel, suspended until some
condition o~urs, or ordered for execution in some way.
The main interpreter loop for this system is identi-
cal to the one discussed previously. However, as the
system is embedded in a dynamic environment, the
procedures appearing in the interpreter must be fast
enough to satisfy the real-time demands placed Ul~On
the system. One way of tailoring and thus improving
the process of option generation is to insert an addi-
tional procedure, post-intention-status, at the end
of the interpreter loop. The purpose of this procedure
is to delay posting events on the event queue regard-
ing any changes to the intention structure until the
end of the interpreter loop. By posting appropriate
events to the event queue the procedure can determine,
among other things, which changes to the intention
structure will be noticed by the option generator. In
this way, one can model various notions of commitment
that result in different behaviours of the agent(ltao and
Georgeff 1992).
Applications
In this section, we consider an air-traffic management
system, OASIS, and relate it to the theoretical formal-
4This is an efficient way of capturing all the paths of
intention-accessible worlds. In other words, the interpreter
does a lazy generation of all possible sequences of actions
that it can intend from the plan library.
ism and the abstract architecture of the previous sec-
tions. The system architecture for OASIS is made up
of one aircraft agent for each arriving aircraft and a
number of global agents, including a sequencer, wind
modeller, coordinator, and trajectory checker.
At any particular time, the system will comprise up to
seventy or eighty agents running concurrently, sequenc-
ing and giving control directives to flow controllers on
a real-time basis. The aircraft agents are responsible
for flying the aircraft and the global agents are respon-
sible for the overall sequencing and coordination of the
aircraft agents. A detailed description of the system
can be found elsewhere (Ljungberg and Lucas 1992).
The system is currently undergoing parallel evaluation
trials at Sydney airport, receiving live data from the
radar.
Modelling: An aircraft agent is responsible for
flying along a certain flight path given by the coor-
dinates of a sequence of waypoints. An example of
the chance or uncertainty in the domain is the wind
field. If this were the only environmental variable, for
each value of the wind velocity at a particular way-
point we would have a corresponding belief-accessible
world. The choices available to an aircraft agent in-
clude flying along various trajectories between its min-
imum speed and maximum speed and at an altitude
between its minimum and maximum altitude. This
can be represented by multiple branches in each of the
belief-accessible worlds mentioned above. As the fi-
nal waypoint is the destination airport, the paths de-
sired by the aircraft agent are those paths where the
calculated ETA of the end node is equal to the de-
sired ETA. The desire-accessible worlds can be ob-
tained from the belief-accessible worlds by pruning
those paths that do not satisfy-the above condition.
The intention-accessible worlds ~an be obtained from
the desire-accessible paths by retaining only ’those that
are the "best" with respect to fuel consumption, air-
craft performance, and so on.
Decision Theory and Commitment: The pri-
mary objective of the sequencer agent is to land all
aircraft safely and in an optimal sequence. Given the
performance characteristics of aircraft, desired separa-
tion between aircraft, wind field, runway assignment,
and a cost function, the sequencing agent uses a num-
ber of different deliberation strategies to compute the
"best" arrival sequence for aircraft and their respec-
tive ETA~s. On determining a particular schedule, the
scheduling agent then single-mindedly commits to the
intention; in other words, the scheduling agent will stay
committed until (a) it believes that all aircraft have
landed in the given sequence; or (b) it does not be-
lieve that there is a possibility that the next aircraft
will meet its assigned ETA. Note that this is not the
classical decision-theoretic viewpoint--any change in
wind field, for example, should, in that view, cause a
recalculation of the entire sequence, even if all aircraft
could still meet their assigned ETAs.
Rao 317
From: Proceedings of the First International Conference on Multiagent Systems. Copyright © 1995, AAAI (www.aaai.org). All rights reserved.

Abstract Interpreter: In the implemented version
of OASIS, each agent in the system deals only with
current beliefs and desires and the options available to
the agent to achieve its desires are written as plans. For
example, although there may be many different ways
of achieving the desired ETA (e.g., flying low at full
speed), the plans of the aircraft agents only include
as options thosetrajectories that are maximally fuel
efficient.
In addition to the above application, PRS and
dMARS have been used in a number of other large-
scale applications, including a system for space shut-
fie diagnosis {Ingrand et al. 1992), telecommunica-
tions network management (lngrand et al. 1992), air-
combat modelling (Rao et al. 1992), and business pro-
cess management. This experience leads us to the firm
conviction that the agent-oriented approach is partic-
ularly useful for building complex distributed systems
involving resource-bounded decision-making.
Essential Features: The essential characteristics
which have contributed to the success of our approach
can be summarized as follows:
¯ The ability to construct plans that can react to spe-
cific situations, can be invoked based on their pur-
pose, and are sensitive to the context of their invo-
cation facilitates modular and incremental develop-
ment. It allows users to concentrate on writing plans
for a subset of essential situations and construct
plans for more specific situations as they debug the
system. As plans are invoked either in response to
particular situations or based on their purpose, the
incremental addition of plans does not require mod-
ification to other existing plans.
¯ The balance between reactive and goal.directed be-
haviour is achieved by committing to plans and pe-
riodically reconsidering such committed plans. The
management of such real-time and concurrent activ-
ities is done by the system, while still giving the user
control in terms of specifying to the system how the
balance is to be achieved. As a result, end-users
need not be involved in complex low-level program-
ming (a difficult and error-prone activity, even for
systems programmers), leading to a reliable system.
¯ The high-level representational and programming
language has meant that end-users can encode their
knowledge directly in terms of basic mental atti-
tudes without needing to master the programming
constructs of a low-level language. This has led
to greater flexibility and shorter development cycles.
For example, when FORTRAN rules that modeUed
pilot reasoning were replaced with plans, the turn-
around time for changes to tactics in an alr-coinbat
simulation system (Rao et al. 1992) improved from
two months to less than a day.
318 ICMAS-9$
Comparison and Conclusion
The BDI architecture draws its inspiration from the
philosophical theories of Bratman (Bratman 1987),
who argues that intentions play a significant and dis-
tinct role in practical reasoning and cannot be reduced
to beliefs and desires. Cohen and Levesque (Cohen
and Levesclue 1990) provided one of the first logical
formalizations of intentions and the notion of commit-
ment. Later formalications include the representation-
alist theory by Konolige and Pollack (Konolige and
Pollack 1993) and the work by Singh (Singh and Asher
1990).
While the earlier formalisms present a particular set
of semantic constraints or axioms as being the formal-
ization of a BDI agent, we adopt the view that one
should be able to choose an appropriate BDI system
for an application based on the rational behaviours re-
quired for that application. As a result, following the
modal logic tradition, we have discussed how one can
categorize different combinations of interactions among
beliefs, desires, and intentions.
A number of agent-oriented systems have been built
in the past few years (Burmeister and Sundermeyer
1992; Georgeff and Lansky 1986; Muller et al. 1994;
Shoham 1993). While many of these appear interest-
ing and have different strengths and weaknesses, none
has yet been applied to as wide a class of complex ap-
plications as the ones discussed in this paper.
Currently, there is very little work on bridging the
gap amongn theory, systems, and applications. The
work by Bratman el. al. (Bratman et al. 1988) de-
scribes the different modules of a BDI architecture
and discusses the philosophical foundations for each of
these modules. However, compared to our abstract in-
terpreter, this model is at a higher level of abstraction
and is not useful as a practical system. More recent
work by Fisher (Fisher 1994) on Concurrent Metatem
specifies agent behaviours as temporal logic specifica-
tions that are directly executed by the system. How-
ever, for applications in which the environment changes
at rates comparable with the calculation cycle of the
system, such theorem provers are unsuited as system
implementations.
The primary contribution of this paper is in inte-
grating the various aspects of BDI agent research--
theoretical foundations from both a quantitative
decision-theoretic perspective and a symbolic rational
agency perspective, system implementation from an
ideal theoretical persepctive to a more practical per-
spective, and the applications that rely on the theoret-
ical foundations and are implemented using a practical
BDI architecture.
Acknowledgements: Tiffs research was supported
by the Cooperative Research Centre [’or Intelligent De-
cision Systems under the Australian Government’s Co-
operative Research C~ntres Program. We also thank
Liz Sonenberg for her valuable comments or! the paper.
From: Proceedings of the First International Conference on Multiagent Systems. Copyright © 1995, AAAI (www.aaai.org). All rights reserved.

References
M. E. Bratman, D. Israel, and M. E. Pollack. Plans
and resource-bounded practical reasoning. Computa-
tional Intelligence, 4:349-355, 1988.
M. E. Bratman. Intentions, Plans, and Practical Rea-
son. Harvard University Press, Cambridge, MA, 1987.
B. Burmeister and K. Sundermeyer. Cooperative
problem-solving guided by intentions and perception.
In E. Werner and Y. Demazeau, editors, Decentralized
A.I. ,9, Amsterdam, The Netherlands, 1992. North
Holland.
P. 1%. Cohen and H. J. Levesque. Intention is choice
with commitment. Artificial Intelligence, 42(3), 1990.
J. Doyle. Rationality and its roles in reasoning. Com-
putational Intelligence, 8(2):376-409, 1992.
E. A. Emerson. Temporal and modal logic. In J. van
Lseuwen, editor, Handbook of Theoretical Computer
Science: Volume B, Formal Models and Semantics,
pages 995-1072. Elsevier Science Publishers and MIT
Press, Amsterdam and Cambridge, MA, 1990.
Michael Fisher. Representing and executing agent-
based systems. In Intelligent Agents: Theories, Ar-
chitectures, and Languages. Lecture Notes in Artifi-
cial Intelligence LNAI 890, Springer Verlag, Amster-
dam, Netherlands, 1994.
M. P. Georgeff mad A. L. Lansky. Procedural knowl-
edge. In Proceedings of the IEEE Special Issue on
Knowledge Representation, volume 74, pages 1383-
1398, 1986.
M. P. Georgeff and A. S. Rao. The semantics of inten-
tion maintenance for rational agents. In Proceedings
of the Fourteenth International Joint Conference on
Artificial Intelligence (IJCAI-95), Montreal, Canada,
August, 1995.
F. F. Ingrand, M. P. Georgeff, and A. S. Rao. An ar-
chitecture for real-time reasoning and system control.
IEEE Expert, 7(6), 1992.
N. R. Jennings. On being responsible. In Y. De-
mazeau and E. Werner, editors, Decentralized’A.I. 3.
North Holland, Amsterdam, Netherlands, 1992.
D. Kinny and M. P. Georgeff. Commitment and ef-
fectiveness of situated agents. In Proceedings of the
Twelfth International Joint Conference on Artificial
Intelligence (IJCAI-91), pages 82-88, Sydney, Aus-
tralia, 1991.
D. Kinny, M. Ljungberg, A. S. Rao, E. A. Sonenberg,
G. Tidhar, and E. Werner. Planned team activity.
In Artificial Social Systems, Lecture Notes in Arti-
ficial Intelligence (LNAI-830), Amsterdam, Nether-
lands, 1994. Springer Verlag.
K. Konolige and M. Pollack. A representationalist
theory of intention. In Proceedings of the Thirteenth
International Joint Conference on Artificial Intelli-
gence (IJCAI-93), Chamberey, France, 1993.
M. Ljungberg and A. Lucas. The OASIS air-traffic
management system. In Proceedings of the Second Pa-
cific Rim International Conference on Artificial Intel-
ligence, PRICA[ ’9~, Seoul, Korea, 1992.
J. Muller, M. Pischel, and M. Thiel. A pragmatic ap-
proach to modelling autonomous interacting systems:
Preliminary report. In Intelligent Agents: Theories,
Architectures, and Languages. Lecture Notes in Arti-
ficial Intelligence LNAI 890, Springer Verlag, Ams-
terdam, Netherlands, 1994.
A. S. Rao and M. P. Georgeff. Asymmetry thesis
and side-effect problems in linear time mad branching
time intention logics. In Proceedings of the Twelfth In-
ternational Joint Conference on Artificial Intelligence
(IJCAI-91}, Sydney, Australia, 1991.
A. S. Rae and M. P. Georgeff. Deliberation and its
role in the formation of intentions. In Proceedings of
the Seventh Conference on Uncertainty in Artificial
Intelligence (UA 1-91). Morgan Kaufmann Publishers,
San Mateo, CA, 1991.
A. S. Rao and M. P. Georgeff. Modelling ratio-
nal agents within a BDI-architecture. In J. Allen,
R. Fikes, and E. Sandewall, editors, Proceedings of
the Second International Conference on Principles
of Knowledge Representation and Reasoning. Morgan
Kaufmnnn Publishers, San Mateo, CA, 1991.
A. S. Rao and M. P. Georgeff. An abstract architec-
ture for rational agents. In C. Rich, W. Swartout,
and B. Nebel, editors, Proceedings of the Third Inter-
national Conference on Principles of Knowledge Rep-
resentation and Reasoning. Morgan Kaufmann Pub-
lishers, San Marco, CA, 1992.
A. S. Rao and M. P. Georgeff. A model-theoretic
approach to the verification of situated reasoning sys-
tems. In Proceedings of the Thirteenth International
Joint Conference on Artificial Intelligence (IJCAl-
93), Chamberey, France, 1993.
A. Rao, D. Morley, M. Selvestrel, and G. Murray.
Representation, selection, and execution of team tac-
tics in air combat modelling. In A. Adams and L. Ster-
ling, editors, Proceedings of the Fifth Australian Joint
Conference on Artificial Intelligence, pages 185-190.
World Scientific, 1992.
S. J. Rosenschein and L. P. Kaelbling. The synthesis
of digital machines with provable epistemic proper-
ties. In J. Y. Halpern, editor, Proceedings of the First
Conference on Theoretical Aspects of Reasoning about
Knowledge, Morgan Kaufmann Publishers, San Ma-
tea, CA, 1986.
Y. Shoham. Agent-oriented programming. Artificial
Intelligence, 60{1):51-92, 1993.
M. Singh and N. Asher. Towards a formal theory
of intentions. In J. van Eijck, editor, Logics in AI
LNAI ~78, pages 472-486. Springer Verlag, Amster-
dam, Netherlands, 1990.
From: Proceedings of the First International Conference on Multiagent Systems. Copyright © 1995, AAAI (www.aaai.org). All rights reserved.