INTELLIGENT MOBILE MULTI-ROBOTIC SYSTEMS: SOME

CHALLENGES AND POSSIBLE SOLUTIONS

avio S. Corr

ea da Silva, Renata Wassermann, Ana Cristina V. Melo,

Leliane N. Barros, Marcelo Finger

Dept. of Computer Science, University of S

ao Paulo

Rua do Mat

ao, 1010, S

ao Paulo, BRAZIL

Keywords:

Distributed multi-agent robotic systems, software for robotic agents, robotic surveillance and security.

Abstract:

Intelligent mobile multi-robotic systems (IMMRSs) are coordinated systems of autonomous mobile robots

endowed with reasoning capabilities. This sort of systems requires the integrated application of a variety of

state-of-the-art techniques developed within the realm of Artiﬁcial Intelligence, as well as instigates the further

development of different specialisations of Artiﬁcial Intelligence. In the present article we examine some

of these techniques and specialisations, discuss some speciﬁc challenges proposed to the ﬁeld of Artiﬁcial

Intelligence by IMMRSs, and suggest possible solutions to these challenges. In order to make our presentation

more concrete, we employ throughout the article a speciﬁc example of IMMRS application, namely security

surveillance of an empty building by a team of robots.

1 INTRODUCTION

Intelligent mobile multi-robotic systems (IMMRSs)

are coordinated systems of autonomous mobile robots

endowed with reasoning capabilities. These systems

require the integrated application of a variety of state-

of-the-art techniques developed within the realm of

Artiﬁcial Intelligence (AI), as well as instigates the

further development of different specialisations of AI.

In the present article we examine some of these

techniques and specialisations, discuss some speciﬁc

challenges proposed to AI by IMMRSs, and sug-

gest possible solutions to these challenges. In order

to make our presentation more concrete, we employ

throughout the article a speciﬁc example of IMMRS

application, namely security surveillance of an empty

building by a team of robots (Gerkey et al., 2004):

Given an environment, modelled as a connected polygonal

free space, a ﬁxed number of searchers – autonomous mo-

bile robots equipped with cameras, each camera having a

ﬁxed angular aperture – and an unknown number of evaders

– autonomous entities which can move arbitrarily fast – de-

termine trajectories for each of the searchers so that the de-

tection of all evaders is guaranteed.

This problem has high computational complex-

ity with respect to the complexity of the environ-

ment (e.g. characterised by the number of edges of

the polygon that comprises it) and the number of

searchers. Moreover, determining the optimal (i.e.

minimum) quantity of searchers given an environment

is NP-hard (Gerkey et al., 2004).

This problem, however, does not take into account

a few parameters that can be important to improve its

accuracy for practical applications. In the following

sections we analyse this problem under the light of

different specialisations of AI, which provide us with

conceptual tools to reﬁne its description. The present

work can be thus regarded as a reﬁnement of the work

presented in (Gerkey et al., 2004), aiming at taking it

from foundational research that provides some math-

ematically well founded guidelines for IMMRS to a

de facto applied work that can effectively be used to

build IMMRS.

In section 2 we add to the model the inherent un-

certainties of the estimates of where a searcher and an

evader are at a given moment. In section 3 we con-

sider the fact that sometimes the sensed environment

and its corresponding model may disagree, requir-

ing a revision of the model of the environment. The

reﬁnements of multi-robotic models that result from

taking into account the features discussed in these two

sections can lead to high consumption of computa-

tional resources. In section 4 we focus on the efﬁ-

ciency of inference systems, to counterbalance this

consumption of resources. In section 5 we consider

the possibility of the searchers communicating with

479

S. Corrêa da Silva F., Wassermann R., Cristina V. Melo A., N. Barros L. and Finger M. (2005).

INTELLIGENT MOBILE MULTI-ROBOTIC SYSTEMS: SOME CHALLENGES AND POSSIBLE SOLUTIONS.

In Proceedings of the Second International Conference on Informatics in Control, Automation and Robotics - Robotics and Automation, pages 479-485

DOI: 10.5220/0001190904790485

 SciTePress

each other, to act in a coordinated fashion, exchang-

ing not only data but also functionalities. In section

6 we discuss research on high level languages to rep-

resent and reason about robotic actions, plans and ca-

pabilities. Finally, in section 7 we present some ﬁnal

discussion and proposed future work.

2 MANAGEMENT OF

UNCERTAIN LOCATION

ESTIMATES

In (Gerkey et al., 2004) the information about the po-

sitions of the searchers and the sensed positions of

the evaders is assumed to be perfectly certain, thus

not taking into account the imprecision of sensors.

Among the many ﬁelds within AI to which uncertain

reasoning is relevant, IMMRSs are perhaps the most

prominent, due to the impossibility to idealise the

environment in which the designed reasoning agents

(autonomous robots in this case) shall inhabit (Gasos

and Safﬁotti, 1999; Safﬁotti, 1999).

Efﬁcient probabilistic reasoning has challenged the

theorists for many years (Dix et al., 2000; Ng and

Subrahmanian, 1992). The usual approach to control

the computational complexity of probabilistic reason-

ing has been to look for special instances of reasoning

problems which present useful probabilistic proper-

ties (Campos and Cozman, 2004; Ide and Cozman,

2004; Rocha and Cozman, 2003).

Let us consider the surveillance problem with a sin-

gle searcher. Differently from the classical formula-

tion of this problem, let us also consider that the ac-

curacy of a sensor varies depending on the positions

of the searcher and the target point in the environment

(e.g. the reliability of the readings of the sensors can

be inversely proportional to the distance separating

the searcher and a point in the environment). Hence,

assuming that positions are characterised by plane co-

ordinates (x, y), if the searcher is at point (x

, y

) and

sends a message stating that there is an evader at point

(x, y), we should from this message infer a collection

of probabilities related to points surrounding (x1, y1)

such that (x

, y

) : µ

should be read as “there is a

probability µ

that point (x

, y

) is occupied by an

evader”.

We must rely on the ability of the searcher to locate

itself in the environment, and the message sent by the

searcher in this setting must contain two pairs of co-

ordinates, one for the actual position of the searcher

, y

) and another for the position of the point pos-

sibly occupied by an evader (x, y). Both pairs of coor-

dinates can be inaccurate, hence if we get a message

from the searcher stating that it is at point (x

, y

we should from this infer a collection of probabil-

ities related to points surrounding (x

, y

) such that

, y

) : µ

should be read as “there is a probabil-

ity µ

that the searcher is at point (x

, y

)”.

The composition of these two probability estimates

can provide us with estimates for the probabilities

that effectively there is an evader at certain points

within the building. Assuming that the plane on

which the searcher navigates is organised as a homo-

geneous square grid, the more accurate the sensors

are, the smaller are the areas with signiﬁcant probabil-

ities given a position (x

, y

) and the areas with sig-

niﬁcant probabilities given a suspicious point (x, y).

Thus, the accuracy of the sensors and the coarseness

of the grid determine the amount of data needed to

estimate probabilities of having evaders at different

points. Accurate sensors and coarser grids make for

smaller amounts of data. Hence, given the accuracy of

the sensors of a searcher, one can control the size of

the lists by making the grid squares larger or smaller.

This work is detailed in (Silva, 2005). An applica-

tion to robot navigation for surveillance is also out-

lined in that reference.

3 EFFICIENT REVISION OF

BELIEFS ABOUT THE

ENVIRONMENT

In (Gerkey et al., 2004) it is assumed that the informa-

tion about the environment is perfectly reliable. Sup-

pose, however, that there are two searchers looking

after an environment. Each searcher receives a map

of the environment, and then the searchers build an

internal model of the environment. As they go around

the physical space, they may observe facts that con-

tradict the information on the map. The robots must

be able to deal with this new information. There are

two main issues involved: (1) is the observation cor-

rect, so that it should be incorporated? (2) If a robot

decides to accept the information, how does it accom-

modate it?

Belief revision (G

ardenfors, 1988; Hansson, 1999)

is the sub-area of AI that deals with these problems.

We are still far from having an acceptable framework

to deal with multiple agents, dynamic worlds and real-

time reasoning. In the present project, we have been

working mainly on the last issue, i.e., how to employ

logical models of belief revision to realistic problems

in which computation time matters.

The main idea is to cope with the high complexity

of logical reasoning by looking for approximate an-

swers. We have extended Cadoli and Schaerf’s frame-

work for approximate reasoning (Sch95) in (Finger

and Wassermann, 2004; Finger and Wassermann,

2005) as a possible solution. In previous work

(Was99; Was01; Chopra et al., 2001), we had al-

ready proposed the use of approximate reasoning in

ICINCO 2005 - ROBOTICS AND AUTOMATION

480

the area of belief revision, but only for the clausal

fragment of propositional logic. Our extension deals

with full propositional logic and has a tableaux-based

proof method that was implemented and tested. We

presented families of anytime reasoners – reasoners

that can be stopped anytime giving an approximate

answer to a query. If the agent is given more time, the

quality of the approximation is improved.

In (Was99), there was another proposal for approx-

imate reasoning, based on the idea of relevance. The

original version was for propositional logic, but has

been extended to ﬁrst-order logic (Ria04a; Ria04b).

Belief revision involving multiple agents is also a

topic which started to receive some attention recently

(Roo03). The main problem here is that when each

agent holds beliefs about other agents’ beliefs, every

new information received by some agent gives rise to

a cascade of actualizations in every agents’ beliefs.

Recent work by Cantwell (Cantwell, 2005) tries to

avoid this problem by deﬁning belief states as prim-

itives. While this works well on the formal side, it

is still not clear how this can be applied to real world

problems, chieﬂy due to computational complexity is-

sues.

4 UNCERTAIN REASONING VIA

APPROXIMATE REASONING

Attributing probability to the conclusion of a logi-

cal reasoner based on the a priori probabilities of the

premises has been for a long time an active area of

research (Nilsson, 1986).

The problem with this approach is its intractabil-

ity, whose sources are both the intractability of logi-

cal inferences and the multiplicity of states that have

to be considered to compute the interval of probabili-

ties of conclusions. The latter problem may be linked

to the fact that the attribution of probabilities to for-

mulae is not truth functional, in the sense that one

cannot always infer the probability of a formula sim-

ply by knowing the probability of its components. So

one has to consider all the exponentially many pos-

sible valuations of a formula to compute probability

intervals, even if the probability of all atomic facts

are known.

Our goal is to solve some of these problems

by using approximations of classical inference, as

introduced by Schaerf and Cadoli (Sch95) and

Dalal (Dalal, 1996). The idea here is to work in some

subclassical logic that proves less theorems than clas-

sical logic, but in which inference is tractable. This

approach was initially restricted to clausal form logic,

without an established proof theory. These approx-

imations have been extended to full classical logic

with a well-deﬁned proof theory (Massacci, 1997;

Finger and Wassermann, 2002; Finger and Wasser-

mann, 2005) and recently, a more reﬁned control on

the complexity of such approximations was estab-

lished (Finger, 2004).

In this work, we expect to apply this latter approach

to deal with some aspects of probabilistic logic. We

assume that we are dealing with a propositional lan-

guage based on a ﬁnite number of atoms p

, . . . , p

and the usual connectives ¬, ∧ , ∨ and →.

The attribution of probabilities to a formula can-

not be inferred from the probabilities of its subfor-

mulae. All we have is that, if P (A) is the attribut-

ion of probabilities to a propositional formula A, such

that 0 ≤ P (A) ≤ 1, then the following must hold:

(1) if ⊢ A then P (A) = 1; (2) if ⊢ ¬(A ∧ B) then

P (A ∨ B) = P (A) + P (B). From this basic fact, it

is possible to infer the following properties: if the

symbol ‘⊢’ represents classical logical consequence,

then: (1) P (¬A) = 1 − P (A); (2) if A ⊢ B then

P (A) ≤ P (B); (3) if ⊢ A ↔ B then P (A) = P (B);

(4) P (A ∨ B) = P (A) + P (B) − P (A ∧ B).

The idea is now to consider ⊢ not as classical infer-

ence, but some tractable approximate inference, and

assume the same deﬁnition for the attribution of prob-

abilities. We want to investigate which of the proper-

ties above are preserved.

We present the tractable approximation here only

in semantic terms; a full proof theory is discussed

in (Finger, 2004). This semantics is called Limited

Bivaluation, LB, and is paramererized by a set Σ of

formulae.

The semantics of LB(Σ) is based on a three-level

lattice, L = (L, ⊓, ⊔, 0, 1), where L is a countable set

of elements L = {0, 1, ǫ

, ǫ

, . . .} such that 0 ⊑ ǫ

⊑

1 for every i < ω and ǫ

6⊑ ǫ

for i 6= j. The ǫ

’s are

called neutral truth values. This lattice is enhanced

with a converse operation, ∼, deﬁned as: ∼ 0 = 1,

∼1 = 0 and ∼ǫ

= ǫ

for all i < ω.

A limited valuation is a function v

: P → L that

maps formulae to elements of the lattice that are sub-

ject to a set of restrictions with regards to whether

a formula is or is not in the parameter set Σ. Ini-

tially, the limited valuation v

maps atoms to the el-

ements of the lattice and is extended to all formu-

lae in the following way: (1) v

(¬A) =∼ v

(A)

(2) v

(A ∧ B) = v

(A) ⊓ v

(B) (3) v

(A ∨ B) =

(A) ⊔ v

(B) (4) v

(A → B) = 1 if v(A) ⊑ v(B); ∼

(A) ⊔ v

(B) otherwise.

A further constraint, called the Limited Bivalence

Restriction, is imposed on v

for formulae in Σ: for-

mulae in Σ are bivalent, that is, if A ∈ Σ then

(A) must be bivalent, that is, v

(A) must satisfy

the rules above for unlimited valuations and be such

that v

(A) = 0 or v

(A) = 1.

When A 6∈ Σ, v

(A) is not always compositional,

which means that a neutral value may be assigned to

A independently of the truth value of its components.

INTELLIGENT MOBILE MULTI-ROBOTIC SYSTEMS: SOME CHALLENGES AND POSSIBLE SOLUTIONS

481

This is the case so that the bivalence of A ∈ Σ can al-

ways be satisﬁed without forcing all A’s subformulae

to be bivalent.

If A ∈ Σ it is always possible to have v

(A) ∈

{0, 1} by making for every atom p in A, v

(p) ∈

{0, 1}. However, this is not the only possibility. For

example, if B, C 6∈ Σ then we can make v

(B) = ǫ

= v

(C), so that v

(B ∧ C) = 0; similarly, we

obtain v

(B ∨ C) = 1 and v

(B → C) = 1.

A valuation v

satisﬁes A if v

(A) = 1, and A is

called satisﬁable; a set of formulae Γ is satisﬁed by

if all its formulae are satisﬁed by v

. A valuation

contradicts A if v

(A) = 0; if A is neither satis-

ﬁed nor contradicted by v

, we say that v

is neutral

with respect to A. A valuation is classical if it assigns

only 0 or 1 to all proposition symbols, and hence to

all formulae.

The notion of a parameterised LB-Entailment,

, is obtained by deﬁning, for a set of formulae Γ

and a formula A, Γ |=

A if no valuation v

such that

(Γ) = 1 also makes v

(A) = 0. This consequence

relation is not classic, for if Γ |=

A and v

(Γ) = 1 it

is possible that A is either neutral or satisﬁed by v

We now deﬁne the tractable entailment |=

, para-

meterized by an integer k. For that, let S be a set

of sets of atoms and, for every Π ∈ S, let Π

the closure of Π under formula formation. We de-

ﬁne Γ |=

A iff there exists a set Π ∈ S such that

Γ |=

A. We deﬁne S

= {Π ⊆ P| |Π| = k}. That is,

is a set of sets of atoms of size k. Note that if we re-

strict our attention to n atoms, |S

| =

= O(n

)

sets of k atoms.

We write |=

to mean |=

. We have focused on

the used of a quadratic decision procedure, |=

. We

then apply the deﬁnition of probability attibution us-

ing such sub-classical entailment, as follows: (i) if

A then P (A) = 1; (ii) if |=

¬(A ∧ B) then

P (A ∨ B) = P (A) + P (B). From that, we can

prove that the following classical properties are pre-

served: if the symbol ‘⊢’ represents classical logi-

cal consequence, then (1) P (¬A) = 1 − P (A); (2) if

A |=

B then P (A) ≤ P (B); (3) if |=

A ↔ B then

P (A) = P (B).

In fact, the results above hold for any integer k, and

not only for k = 2. However, the following property

fails: P (A ∨ B) = P (A) + P (B) − P (A ∧ B). This is

due to the fact that a classical theorem does not hold,

in general, for any k: 6|=

A ∨ B ↔ (A ∧ ¬B) ∨ (A ∧

B) ∨ (¬A ∧ B). In fact, to obtain an expression for

P (A ∨ B) we have to consider the cases where A and

B can have neutral values ǫ

That is, although it may be less complex to com-

pute an inference in |=

than classical inference, the

computation of probabilities has to take in considera-

tion many more terms.

Future work consider the investigation of the com-

plexity of using probabilistic logic based on |=

and

the effects of loss of expressivity on this inference in

the computation of probabilities.

5 FORMAL SPECIFICATION

AND VERIFICATION OF

IMMRSS

As usual, it is assumed in (Gerkey et al., 2004) that

the embedded software in each robot is ﬁxed within

that robot. Indeed, it is usually accepted that each ro-

bot is the physical embodiment of an agent in a mul-

tiagent system. To our understanding, this constraint

in the design of IMMRSs is unnecessary and restric-

tive. If robots can exchange messages to coordinate

their actions, there is no reason why they cannot also

exchange lines of code that implement actions them-

selves. This amounts to a decoupling of the notions

of robot and agent: a single robot, in this framework,

can accomodate more than one agent simultaneously,

and agents can migrate between robots.

The decoupling of the notions of robot and agent

greatly extends the ﬂexibility in the speciﬁcation, de-

sign and implementation of IMMRSs. It also makes

these activities far more complex. In order to keep

such systems under control, it can be important to em-

ploy formal techniques for the speciﬁcation and veri-

ﬁcation of mobile agents for robotic systems.

The π-calculus (Milner, 1999) is a theory for

mobile agents based on an algebra for concurrent

processes, the CCS (Milner, 1989). In a very simpli-

ﬁed way, mobile agents can be regarded as concurrent

processes endowed with the capability of dynamic re-

conﬁguration. The π-calculus has added to the CCS,

among other things, features of mobility and dynamic

reconﬁguration.

It is relevant for the formal speciﬁcation and ver-

iﬁcation of mobile agents the development of tools

for automatic veriﬁcation of mobile agents, based on

novel veriﬁcation techniques or the combination of

existing techniques. Our work has focused on (1)

techniques to get rid of useless code in the speciﬁ-

cation of mobile agents and their formal speciﬁcation;

(2) novel techniques for the formal veriﬁcation of mo-

bile agents; (3) combination of formal veriﬁers of mo-

bile agents; (4) ontologies for mobile agents and mo-

bility features, to support the combination of systems

for speciﬁcation and veriﬁcation of mobile agents; (5)

ontologies about the capabilities of formal veriﬁers;

(6) formal models of autonomous and mobile agents;

(7) automatic code generation for the communication

between mobile agents; and (8) tools for the speciﬁ-

cation and integration of formal veriﬁers for mobile

agents.

ICINCO 2005 - ROBOTICS AND AUTOMATION

482

The behaviour of mobile agents must be tested,

based on explicitly deﬁned criteria (Wey88; McGre-

gor and Korson, 1994; Chen and Kao, 1999; Kung

et al., 1995) that determine what should be tested.

Our work on this ﬁeld has focused on (1) the study

of what sets of properties can be veriﬁed in programs,

and how the veriﬁcation of these properties simpliﬁes

the tests requirements; (2) tools to support the com-

bination of formal veriﬁcation and tests, e.g. the au-

tomation of object control graphs; (3) the extension

of model based formal methods to the speciﬁcation of

exception handling; (4) the study of how to use for-

mal veriﬁers to improve the testing of software com-

ponents with exception handling.

Some preliminary results in these topics were pub-

lished in (Melo, 2005; Melo, 2004; Nunes and Melo,

2004; Andrade et al., 2004; Mel04b; Melo, 2003).

6 HIGH LEVEL REASONING

ABOUT ROBOTIC ACTIONS

The problem solved in (Gerkey et al., 2004) is how

to determine the plans (i.e. trajectories in the environ-

ment) for a collection of searchers, based on primitive

actions such as move forward and turn 90 degrees to

the left. This formulation of the problem scales badly

as the number of searchers increases. Another possi-

ble planning strategy is a higher-level set of actions,

possibly involving teams of robots, speciﬁed in terms

of the above primitive actions. This is an AI planning

approach called Hierarquical Task Network Planning

(HTN), that has been proved to be more powerful then

the search in the state/action space solution using only

primitive actions (Kutluhan et al., 1995).

The decomposition of a goal task adds other goal

tasks to be decomposed and this is very similar to

what it is done in a regressive partial-order planner

that adds actions to satisfy the goal making the pre-

conditions of these actions to become new goals (sub-

goaling) to plan for. In fact, it has been proved (Kut-

luhan et al., 1995) that planning with goal tasks al-

lows the HTN planners to inherit the efﬁciency of the

partial-order planners. HTN planners have been ap-

plied in many applications, such as a system for in-

tegrated product design and manufacturing planning,

and computer games like the winer of the 1997 world

championship of computer bridge.

Golog (Levesque et al., 1997) is a logical language,

implemented as a Prolog meta-interpreter, that allows

the deﬁnition of procedures to describe the behavior

of an agent, using the Situation Calculus (McCarthy,

1963) to represent knowledge, actions and states of

the world. A programmer can specify a robot high-

level control program as a set of high-level proce-

dures, which are decomposed by Golog into low-level

procedures (or primitive actions) during execution

time, very similar to HTN planning. Golog is a high-

level agent programming language, in which standard

programming constructs (e.g. sequence, choice and

iteration) are used to write the agent control program.

It can effectively represent and reason about the ac-

tions performed by agents in dynamic environments.

The emerging success of Golog has shown that, by us-

ing a logical approach, it is possible to solve complex

robotic tasks efﬁciently, despite contrary belief. How-

ever, Golog uses a planning strategy based on situa-

tion calculus, a logical formalism in which plans are

represented as a totally ordered sequence of actions

and, therefore, it inherits the well known deﬁciencies

of this approach (Kutluhan et al., 1995).

Since Golog decomposes procedures in a very si-

miliar way that it is done by HTN planning (Ga-

baldon, 2002), we have been working on the pro-

posal of a number of extensions in the Golog meta-

interpreter (Barros and Iamamoto, 2003; Iamamoto,

2005), namely (1) we have introduced the idea of goal

tasks into the Golog language; (2) based on the for-

malisation of HTN planning (Kutluhan et al., 1995),

we have proposed a special kind of Golog procedures,

called Goal Procedures, that can be decomposed by

the Golog meta-interpreter to solve goals of attain-

ment problems; (3) we have presented a way to in-

terleave Goal Procedures to solve action interactions

(conﬂicting subgoals); (4) we have also compared the

efﬁciency of using Goal Procedures with an example

of a Situation Calculus planner proposed by Reiter,

the Wspbf planner, encoded in Golog (Rei01) and

showed that our proposal, besides being more com-

patible with Golog way of planning through decom-

positions, can be also more efﬁcient; (5) we have ar-

gued that Goal Procedures can be very useful in an

on-line execution to replanning when action’s execu-

tion fails.

We are currently working on the use of Goal Pro-

cedures to replanning in IndiGolog, an extention of

GOLOG to perform on-line execution of programs.

We have also proposed a new high-level robot

programming language, called ABGOLOG (Per04b;

Per04c; Per04a). This language is based on Golog,

but it uses Event Calculus as the formalism to de-

scribe actions and abductive reasoning to synthesize

plans, which corresponds to partial order planning.

So, based on our previous work on implementation

and analysis of abductive event calculus planning sys-

tems (Per04b), we have shown how it is possible to

modify AbGolog’s implementation to improve its ef-

ﬁciency, according to speciﬁc domain characteristics.

In (Tre05) we have shown an example of a robot

application using a Lego



MindStorms

robot. The

implementation was done in Legolog, a package com-

posed by the Indigolog language and communication

protocols for the Lego



MindStorms

robot.

INTELLIGENT MOBILE MULTI-ROBOTIC SYSTEMS: SOME CHALLENGES AND POSSIBLE SOLUTIONS

483

7 CONCLUSION

The design and implementation of IMMRSs is a great

challenge to AI and related areas. Chief among the

reasons to make this sort of systems so challenging

is the fact that one cannot idealise the environments

upon which the systems shall act, and hence they re-

quire reﬁned methods to ensure the coupling between

idealised models (based on which the systems are pro-

grammed) and physical environments.

In the present article we listed some areas of Ar-

tiﬁcial Intelligence that we believe are most relevant

to multi-robotic systems, and that at the same time are

more strikingly challenged by those systems. We sug-

gested some solutions to speciﬁc challenges, which

are the ones on which we are working at the moment.

REFERENCES

A. G. Andrade, A. C. V. Melo, and M. M. Amorim. Da

especiﬁcao veriﬁcao de agentes mveis – um ambiente

grﬁco. In Mauricio Solar and David Fernandez Baca,

editors, Anales de 30th Conferencia Latino Americana

de Informtica, Arequipa, Peru. 2004.

L. N. de Barros and E. Iamamoto. Planning for tasks in cog-

nitive robotics. Proceedings of VI Simpsio Brasileiro

de Automao Inteligente, Bauru, SP, Brazil. 2003.

C. P. Campos, and F. G. Cozman. Inference in credal net-

wors using multilinear programming. Second Start-

ing AI Researcher Symposium (STAIRS), 50-61, IOS

Press. 2004.

J. Cantwell. A formal model of multi-agent belief-

interaction. Journal of Logic, Language and Infor-

mation. 2005. to appear.

M.-H. Chen, and H. M. Kao. Testing object-oriented pro-

grams - an integrated approach. In Proceedings of the

Tenth International Symposium on Software Reliabil-

ity Engineering. IEEE Computer Society Press. 1999.

S. Chopra, R. Parikh and R. Wassermann. Approximate

belief revision. Logic Journal of the IGPL, 9(6):755–

768. 2001.

F. S. Correa da Silva. Where am I? Where are you? Tech-

nical Report 2005-03. Department of Computer Sci-

ence, University of Sao Paulo. 2005.

M. Dalal. Anytime families of tractable propositional rea-

soners. In International Symposium of Artiﬁcial Intel-

ligence and Mathematics AI/MATH-96, pages 42–45.

1996.

J. Dix, M. Nanni, and V. S. Subrahmanian. Probabilistic

agent programs. ACM Transactions on Computational

Logic, v. 1(2), 207–245. 2000.

M. Finger. Polynomial approximations of full propositional

logic via limited bivalence. In 9th European Confer-

ence on Logics in Artiﬁcial Intelligence (JELIA 2004),

LNAI vol. 3229, pages 526–538. 2004.

M. Finger and R. Wassermann. Expressivity and control

in limited reasoning. In Frank van Harmelen, editor,

15th European Conference on Artiﬁcial Intelligence

(ECAI02), pages 272–276, Lyon, France. 2002.

M. Finger and R. Wassermann. Approximate and limited

reasoning: Semantics, proof theory, expressivity and

control. Journal of Logic And Computation, 14(2):179

– 204. 2004.

M. Finger and R. Wassermann. The universe of proposi-

tional approximations. Theoretical Computer Science.

2005. to appear.

A. Gabaldon. Programming hierarchical task networks

in the situation calculus. AIPS-2002 Workshop on

On-line Planning and Scheduling. Toulouse, France.

2002.

P. G

ardenfors. Knowledge in Flux - Modeling the Dynamics

of Epistemic States. MIT Press. 1988.

J. Gasos, and A. Safﬁotti. Using fuzzy sets to represent

uncertain spatial knowledge in autonomous robots.

Spatial Cognition and Computation, v. 1, 205-226,

Kluwer. 1999.

B. P. Gerkey, S. Thrun, and G. Gordon. Visibility-based

pursuit-evasion with limited ﬁeld of view. In Proceed-

ings of the National Conference on Artiﬁcial Intelli-

gence (AAAI 2004). AAAI Press. 2004.

S. O. Hansson. A Textbook of Belief Dynamics. Kluwer

Academic Press. 1999.

E. Iamamoto. HTN Planning in Golog. Msc Dissertation at

the Departament of Computer Science of the Univer-

sity of So Paulo, Brazil. 2005.

J. S. Ide, and F. G. Cozman. IPE and L2U: Approximate al-

gorithms for credal networks. Second Starting AI Re-

searcher Symposium (STAIRS), 118-127, IOS Press.

2004.

D. Kung, N. Suchak, P. Hsia, Y. Toyoshima, and C. Chen.

Object state testing for object-oriented programs.

In Proc. of COMPSAC’95. IEEE Computer Society

Press. 1995.

E. Kutluhan, J. Hendler, and D. Nau. Complexity, decidabil-

ity and undecidability results for domain-independent

planning. Artiﬁcial Intelligence 76(1-2):75-88. 1995.

H. J. Levesque, R. Reiter, Y. Lesprance, F. Lin, and R. B.

Scherl. Golog: A logic programming language for dy-

namic domains. 1997.

F. Massacci. A proof theory for tractable approximations of

propositional reasoning. In Maurizio Lenzerini, edi-

tor, AI*IA-97, volume 1321 of LNAI, pages 219–230.

SV. 1997.

J. McCarthy. Situations, actions and causal laws, Techni-

cal report, Stanford University. Reprinted in Semantic

Information Processing (M. Minsky ed.), MIT Press,

Cambridge, MAss., 1968, pp. 410-417. 1963.

J. D. McGregor, and T. D. Korson. Integrated object-

oriented testing and development processes. Commu-

nications of ACM, 37(9). 1994.

A. C. V. Melo. From active names to pi-calculus rewriting

rules. ENTCS (Electronic Notes in Theoretical Com-

puter Science), to appear. 2005.

A. C. V. Melo. π-calculus rewriting rules based on active

names. In Alexandre Mota and Arnaldo Moura, edi-

tors, SBMF2004: 7th Brazilian Symposium on Formal

Methods, Recife, PE, Brazil. 2004.

ICINCO 2005 - ROBOTICS AND AUTOMATION

484

A. C. V. Melo. A study on the potential active names of π-

agents. ENTCS (Electronic Notes in Theoretical Com-

puter Science), 95(C):269–286. 2004.

A. C. V. Melo. A study on the potential active names of π-

agents. In A. L. C. Cavalcanti and Patrcia Machado,

editors, WMF2003: 6th Brazilian Workshop on For-

mal Methods, Campina Grande, PB, Brazil. 2003.

R. Milner. Communication and Concurrency. Prentice–

Hall, ﬁrst edition. 1989.

R. Milner. Communicating and Mobile Systems: the π-

Calculus. Cambridge University Press. 1999.

R. Ng, and V. S. Subrahmanian. Probabilistic logic pro-

gramming. Information and Computation, v. 101(2),

150–201. 1992.

N. Nilsson. Probabilistic logic. Artiﬁcial Intelligence,

28(1):71-87. 1986.

P. R. Nunes and A. C. V. Melo. Ocongra - uma ferramenta

para gerao de grafos de controle de ﬂuxo de obje-

tos. In Murilo Camargo and Jaelson Castro, editors,

Anais do Simpsio Brasileiro de Engenharia de Soft-

ware, Braslia, Brasil. 2004.

S. L. Pereira and L. N. Barros. High-Level Robot Pro-

grams Based on Abductive Event Calculus. Inter-

national Workshop on Cognitive Robotics, CogRob-

2002. 2002.

S. L. Pereira and L. N. Barros. Formalizing planning algo-

rithms: a logical framework for the research on ex-

tending the classical planning approach. International

Conference on Planning Systems Workshop: Con-

necting Planning Theory with Practice. 2004.

S. L. Pereira and L. N. Barros. Planning with Abduction:

a logical framework to explore extensions to classical

planning. Simpsio Brasileiro de Inteligncia Artiﬁcial,

LNAI. 2004.

S. L. Pereira and L. N. Barros. High-level Robot Pro-

gramming: an abductive approach using event calcu-

lus. Simpsio Brasileiro de Inteligncia Artiﬁcial, LNAI.

2004.

R. Reiter. Knowledge in Action: Logical Foundations for

Specifying and Implementing Dynamical Systems.

MIT Press. 2001.

J. Riani. Towards an efﬁcient inference proce-

dure through syntax based relevance. Mas-

ter’s thesis, Department of Computer Science,

University of S

ao Paulo. 2004. Available at

http://www.ime.usp.br/˜joselyto/mestrado.

J. Riani and R. Wassermann. Using relevance to speed up

inference – some empirical results. In Proceedings of

the Brazilian Artiﬁcial Intelligence Symposium, Lec-

ture Notes on Artiﬁcial Intelligence. Springer-Verlag.

2004.

J. C. F. Rocha, and F. G. Cozman. Inference in credal

networks with branch-and-bound algorithms. Third

International Symposium on Imprecise Probabilities

and Their Applications, 482–495, Carleton Scientiﬁc.

2003.

J. W. Roorda, W. Van der Hoek, and J. J. Meyer. Iterated

belief change in multi agent systems. Logic Journal

of the IGPL, 11(2):223–246. 2003.

A. Safﬁotti. Handling uncertainty in control of autonomous

robots. in Wooldridge, M., Veloso, M. (eds.) Artiﬁcial

Intelligence Today, 381–408, Springer-Verlag. 1999.

M. Schaerf and M. Cadoli. Tractable reasoning via approx-

imation. Artiﬁcial Intelligence, 74(2):249–310. 1995.

F. W. Trevizan and L. N. de Barros. Planejamento e Exe-

cuo em Golog para Robs Lego Mindstorm. Encontro

Nacional de Inteligncia Artiﬁcial. 2005.

R. Wassermann. Resource-Bounded Belief Revision. PhD

thesis, Institute for Logic, Language and Computation

— University of Amsterdam. 1999.

R. Wassermann. On structured belief bases. In Hans Rott

and Mary-Anne Williams, editors, Frontiers in Belief

Revision. Kluwer. 2001.

E. J. Weyuker. The evaluation of program-based software

test data adequacy. Communications of ACM, 31(6).

1988.

INTELLIGENT MOBILE MULTI-ROBOTIC SYSTEMS: SOME CHALLENGES AND POSSIBLE SOLUTIONS

485