SEMIOTICS AND HUMAN-ROBOT INTERACTION

ao Silva Sequeira

Institute for Systems and Robotics, Instituto Superior T

ecnico

Av. Rovisco Pais 1, 1049-001 Lisbon, Portugal

Maria Isabel Ribeiro

Institute for Systems and Robotics, Instituto Superior T

ecnico

Av. Rovisco Pais 1, 1049-001 Lisbon, Portugal

Keywords:

Semi-autonomous robot, human-robot interaction, semiotics.

Abstract:

This paper describes a robot control architecture supported on a human-robot interaction model obtained

directly from semiotics concepts.

The architecture is composed of a set of objects deﬁned after a semiotic sign model. Simulation experiments

using unicycle robots are presented that illustrate the interactions within a team of robots equipped with skills

similar to those used in human-robot interactions.

1 INTRODUCTION

This paper describes part of an ongoing project

aiming at developing new methodologies for semi-

autonomous robot control that, in some sense, mimic

those used by humans in their relationships. The pa-

per focus (i) in the use of semiotic signs to control

robot motion and (ii) their use in the interaction with

humans. The approach followed deﬁnes (i) a set of

objects that capture key features in human-robot and

robot-robot interactions and (ii) an algebraic system

with operators to work on this space of the objects.

In a wide variety of applications of robots, such

as surveillance in wide open areas, rescue missions

in catastrophe scenarios, and working as personal as-

sistants, the interactions among robots and humans

are a key factor in the success of a mission. In such

missions contingency situations may lead a robot to

request external help, often from a human operator,

and hence it seems natural to search for interaction

languages that can be used by both. Furthermore, a

uniﬁed framework avoids the development of separate

competences to handle human-robot and robot-robot

interactions.

The numerous robot control architectures proposed

in the literature account for human-robot interac-

tions (HRI) either explicitely through interfaces to

handle external commands, or implicitely, through

task decomposition schemes that map high level mis-

sion goals into motion commands. Examples of cur-

rent architectures developed under such principles can

be found in (Aylett and Barnes, 1998),(Huntsberger

et al., 2003),(Albus, 1987),(Kortenkamp et al., 1999).

If the humans are assumed to have enough knowledge

on the robots and the environment, imperative com-

puter languages can be used for HRI, easily leading to

complex communication schemes. Otherwise, declar-

ative, context dependent, languages, like Haskell, (Pe-

terson et al., 1999) and FROB, (Hager and Peterson,

1999), have been proposed to simulate robot systems

and also as a mean to interact with them. BOBJ

was used in (Goguen, 2003) to illustrate examples on

human-computer interfacing. RoboML, (Makatchev

and Tso, 2000), supported on XML, is an example

of a language explicitly designed for HRI, account-

ing for low complexity programming, communica-

tions and knowledge representation. In (Nicolescu

and Matari

c, 2001) the robots are equipped with be-

haviors that convey information on their intentions to

the outside environment. These behaviors represent a

form of implicit communication between agents such

as the robot following a human without having been

told explicitly to do so. This form of communication,

without explicit exchange of data is an example of in-

teraction that can be modeled using semiotics.

Semiotics studies the interactions among humans.

These are often characterized by the loose speciﬁca-

tion of objectives for instance as when a sentence ex-

presses an intention of motion instead of a speciﬁc

path. Capturing some of these features, typical of nat-

ural languages, is the aim of this project.

The paper is organised as follows. Section 2

Sequeira J. and Ribeiro M. (2006).

SEMIOTICS AND HUMAN-ROBOT INTERACTION.

In Proceedings of the Third International Conference on Informatics in Control, Automation and Robotics, pages 58-65

DOI: 10.5220/0001220600580065

 SciTePress

presents key concepts in semiotics to model human-

human interactions and motivates their use to model

human-robot interactions. These concepts are used

in Section 3 to deﬁne the building blocks of the pro-

posed architecture. Section 4 presents a set of basic

experiments that demonstrate the use of the semiotic

model developed along the paper. Section 5 presents

the conclusions and future work.

2 A SEMIOTIC PERSPECTIVE

FOR HRI

In general robots and humans work at very different

levels of abstraction. Humans work primarily at high

levels of abstraction whereas robots are commonly

programmed to follow trajectories, hence operating at

a low level of abstraction.

Semiotics is a branch of general philosophy which

studies the interactions among humans, such as the

linguistic ones (see (Chandler, 2003) for an introduc-

tion to semiotics) and hence it is a natural frame-

work where to search for new methodologies to model

human-robot interactions. In addition, this should be

extended to robot-robot interactions yielding a unify-

ing framework to handle any interaction between a

robot and the external environment.

An algebraic formulation for semiotics and its use

in interface design has been presented in (Malcolm

and Goguen, 1998). An application of hypertext the-

ory to world wide web was developed in (Neum

uller,

2000) Machine-machine and human-human interac-

tions over electronic media (such as the Web) have

been addressed in (Codognet, 2002). Semiotics has

been use in intelligent control as the term encompass-

ing the main functions related to knowledge, namely

acquisition, representation, interpretation, transfor-

mation and manipulation (see for instance (Meystel

and Albus, 2002)) and often directly related to the

symbol grounding concept of artiﬁcial intelligence.

The idea underlying semiotics is that humans com-

municate among each other (and with the environ-

ment) through signs. Roughly, a sign encapsulates

a meaning, an object, a label and the relations be-

tween them. Sign systems are formed by signs and

the morphisms deﬁned among them (see for instance

(Malcolm and Goguen, 1998) for a deﬁnition of sign

system) and hence, under reasonable assumptions on

the existence of identity maps, map composition, and

composition association, can be modeled using cate-

gory theory framework. The following diagram, sug-

gested by the “semiotic triangle” diagram common

in the literature on this area (see for instance (Chan-

dler, 2003)), expresses the relations between the three

components of a sign.

syntactics



syntactics



semantics

pragmatics





pragmatics



semantics



(1)

Labels, (L), represent the vehicle through which the

sign is used. Meanings, (M), stand for what the users

understand when referring to the sign. The Objects,

(O), stand for the real objects signs refer to.

The morphisms in diagram (1) represent the dif-

ferent perspectives used in the study of signs. Semi-

otics currently considers three different perspectives:

semantics, pragmatics and syntactics, (Neum

uller,

2000). Semantics deals with the general relations

among the signs. For instance, it deﬁnes whether or

not a sign can have multiple meanings. In human-

human interactions amounts to say that different lan-

guage constructs can be interpreted equivalently, that

is as synonyms. Pragmatics handles the hidden mean-

ings that require the agents to perform some infer-

ence on the signs before extracting their real meaning.

Syntactics deﬁnes the structural rules that turn the la-

bel into the object the sign stands for. The starred

morphisms are provided only to close the diagrams

(as the “semiotic triangle” is usually represented as

an undirected graph).

3 AN ARCHITECTURE FOR HRI

This section introduces the architecture by ﬁrst deﬁn-

ing a set of context free objects and operators on this

set that are compatible (in the sense that they can be

identiﬁed) with diagram (1). Next, the corresponding

realizations for the free objects are described. The

proposed architecture includes three classes of ob-

jects: motion primitives, operators on the set of mo-

tion primitives and decision making systems (on the

set of motion primitives).

The ﬁrst free object, named action, deﬁnes primi-

tive motions using simple concepts that can be easily

used in a HRI language. The actions represent mo-

tion trends, i.e., an action represents simultaneously a

set of paths that span the same bounded region of the

workspace.

Deﬁnition 1 (Free action) Let k be a time index, q

the conﬁguration of a robot where the action starts to

be applied and a(q

the conﬁguration at time k of

a path generated by action a.

SEMIOTICS AND HUMAN-ROBOT INTERACTION

A free action is deﬁned by a triple A ≡ (q

,a,B

)

where B

is a compact set and the initial condition of

the action, q

, veriﬁes,

∈ B

, (2)

a(q

= q

, (2b)

∃

>

min

: B(q

,) ⊆ B

, (2c)

with B(q

,) a ball of radius  centered at q

, and

∀

k≥0

a(q

∈ B

. (2d)



Actions are to inclose different paths with similar

(in a wide sense) objectives. Paths that can be con-

sidered semantically equivalent, for instance because

they lead to a successful execution of a mission, may

be enclosed within a single action.

Representing the objects in Deﬁnition 1 in the

form of a diagram it is possible to establish a corre-

spondence between (free) actions and signs verifying

model (1),



semantics

,a,B

)

semantics

synctatics



pragmatics

The projection maps semantics

and semantics

express the fact that multiple trajectories starting in

a neighborhood of q

and lying inside B

may lead

to identical results. The pragmatics

map expresses

the fact that given a bounding region it may be pos-

sible to infer the action being executed. Similarly,

pragmatics

represents the maps that infer the ac-

tion being executed given the initial condition q

. The

synctatics map simply expresses the construction of

an action through Deﬁnition 1.

Following model (1), different actions can yield the

same meaning, that is, the two actions can produce the

same net effect in a mission. This amounts to require

that the following diagram commutes,

equality



semantics





semantics



(3)

where 1

stands for the identity map in the space of

meanings, M.

Diagram (3) provides a roadmap to deﬁne action

equality as a key concept to evaluate sign semantics.

Deﬁnition 2 (Free action equality) Two actions

) and (a

) are equal, the rela-

tion being represented by a

)=a

), if and

only if the following conditions hold

),a

) ⊂ B

∩ B

(4)

∀

≥0

, ∃

≥0

, ∃



∈B(a

,) ⊂ B

∩ B

(4b)



Expressions (4) and (4b) deﬁne the equality map

in diagram (3). It sufﬁces to choose identical bound-

ing regions (after condition (4)) and a goal region

inside the bounding regions (after condition (4b)) to

where the paths generated by both actions converge.

The realization for the free action of Deﬁnition 1 is

given by the following proposition.

Proposition 1 (Action) Let a(q

) be a free action.

The paths generated by a(q

) are solutions of a sys-

tem in the following form,

˙q ∈ F

(q) (5)

where F

is a Lipschitzian set-valued map with closed

convex values verifying,

(q) ⊆ T

(q) (5b)

where T

(q) is the contingent cone to B

at q (see

(Smirnov, 2002) for the deﬁnition of this cone).



The demonstration of this proposition is just a re-

statement, in the context of this paper, of Theorem 5.6

in (Smirnov, 2002) on the existence of invariant sets

for the inclusion (5).



The convexity of the values of the F

map must be

accounted for when specifying an action. The Lip-

schitz condition imposes bounds on the growing of

the values of the F

map. In practical applications this

assumption can always be veriﬁed by proper choice

of the map. This condition is related to the existence

of solutions to (5), namely as it implies upper semi-

continuity (see (Smirnov, 2002), Proposition 2.4).

Proposition 2 (Action identity) Two actions a

and

, implemented as in Proposition 1, are said equal

if,

= B

(6)

∃

: ∀

k>k

(q(k)) = F

(q(k)) (6b)



ICINCO 2006 - ROBOTICS AND AUTOMATION

The demonstration follows by direct veriﬁcation of

the properties in Deﬁnition 2.

By assumption, both actions verify the conditions

in Proposition 1 and hence their generated paths are

contained inside B

∩ B

which implies that (4) is

veriﬁed.

Condition (6b) states that there are always motion

directions that are common to both actions. For ex-

ample, if any of the actions a

generates paths re-

stricted to F

∩ F

then condition (4b) is veriﬁed.

When any of the actions generates paths using motion

directions outside F

∩ F

then condition (6b) indi-

cates that after time k

they will be generated after the

same set of motion directions. Both actions generate

paths contained inside their common bounding region

and hence the generated paths verify (4b).



A sign system is deﬁned by the signs and the mor-

phisms among them. In addition to the equality map,

two other morphisms are deﬁned: action composition

and action expansion.

Deﬁnition 3 (Free action composition) Let a

)

and a

) be two free actions. Given a compact

set M, the composition a

j◦i

)=a

) ◦ a

)

veriﬁes,

if B

∩ B

= ∅

j◦i

) ⊂ B

∪ B

(7)

∩ B

⊇ M (7b)

otherwise, the composition is undeﬁned.



Action a

j◦i

) resembles action a

) up to the

event marking the entrance of the paths into the re-

gion M ⊆ B

∩ B

. When the paths leave the

common region M the composed action resembles

). While in M the composed action generates

a link path that connects the two parts.

Whenever the composition is undeﬁned the follow-

ing operator can be used to provide additional space

to one of the actions such that the overlapping region

is non empty.

Deﬁnition 4 (Free action expansion) Let a

)

and a

) be two actions with initial conditions at

and q

respectively. The expansion of action a

by action a

, denoted by a

)  a

), veriﬁes

the following properties,

ji

= B

∪ B

, with B

∩ B

⊆ M (8)

where M is a compact set representing the minimal

amount of space required for the robot to maneuver

and such that the following property holds

∃

∈B

: a

)=a

) (8b)

meaning that after having reached a neighborhood of

, a

) behaves like a



Given the realization (1) chosen for the actions, the

composition and expansion operators can be realized

through the following propositions.

Proposition 3 (Action composition) Let a

and a

be two actions deﬁned by the inclusions

˙q

∈ F

) and ˙q

∈ F

)

with initial conditions q

and q

, respectively. The

action a

j◦i

) is generated by ˙q ∈ F

j◦i

(q), with the

map F

j◦i

given by

j◦i

⎧

⎪

⎨

⎪

⎩

) if q  B

\M (3)

) ∩ F

) if q ∈ M (3b)

) if q ∈ B

\M (3c)

∅ if B

∩ B

= ∅ (3d)

for some M ⊂ B

∩ B

Outside M the values of F

and F

verify the con-

ditions in Proposition 1. Whenever q ∈ M then

) ∩ F

) ⊂ T

(q).



The ﬁrst trunk of the resulting path, given by (3),

corresponds to the path generated by action a

)

prior to the event that determines the composition.

The second trunk, given by (3b), links the paths gen-

erated by each of the actions. Note that by imposing

that F

) ∩ F

) ⊂ T

) the link paths can

move out of the region M. The third trunk, given

by (3c), corresponds to the path generated by action

By Proposition 1, each of the trunks is guaranteed

to generate a path inside the respective bounding re-

gion and hence the overall path veriﬁes (7).



The action composition in Proposition 3 generates

actions that resemble each individual action outside

the overlapping region. Inside the overlapping area

the link path is built from motion directions common

to both actions being composed. The crossing of the

boundary of M deﬁnes the events marking the transi-

tion between the trunks.

Whenever F

) ∩ F

)=∅ it is still possible

to generate a link path, provided that M has enough

space for maneuvering. The basic idea, presented in

the following proposition, is to locally enlarge either

) or F

Proposition 4 (Action expansion) Let a

and a

two actions deﬁned after the inclusions

˙q

∈ F

) and ˙q

∈ F

)

SEMIOTICS AND HUMAN-ROBOT INTERACTION

The expansion a

ji

) veriﬁes the following prop-

erties

ij



if q  B

\M (9)

∪ F

if q ∈ B

∩ B

∪ M (9b)

where M ⊇ B

∩B

is the expansion set chosen large

enough such that F

∪ F

veriﬁes (5b).



Condition (9) generates paths corresponding to the

action a

). These paths last until an event, trig-

gered by the crossing of the boundary of M, is de-

tected. This crossing determines an event that ex-

pands the overall bounding region by M and the set

of paths, by F

, as expressed by (9b).

Assuming that F

∪ F

⊂ T

∩B

∪M

, that is, it

veriﬁes (5b), the complete path is entirely contained

inside the expanded bounding region. After moving

outside M paths behave as of generated by action a

as required by (8b).



Instead of computing a priori M, the expansion op-

erator can be deﬁned as a process by which action

converges to action a

in the sense that F

) →

) and M is the space spanned by this process.

The objects and operators deﬁned above are com-

bined in the following diagram,

◦





(9)

where A stands for the set of actions available to the

robot, A

◦

and A



stand for the sets of actions rep-

resenting primitive motions used in composition and

expansion, respectively.

The product like part (in the lefthand part) of dia-

gram (9) represents the supervision strategy that de-

termines which of the action sets, A

◦

or A



, is se-

lected at each event. This strategy is mission depen-

dent and is detailed in the s

◦

and s



maps.

4 EXPERIMENTAL RESULTS

This section presents two basic simulation experi-

ments using a team of three unicycle robots and an

experiment with a single real unicycle robot.

Missions are speciﬁed through goal regions the ro-

bots have to reach. This resembles a typical human-

robot interaction, with the human specifying a mo-

tion trend the robot has to follow through the goal

region. Furthermore, intuitive bounding regions are

easily constructed from them. The human operator is

thus not explicitely present in the experiments. Nev-

ertheless, as results clear from the third experiment

(with the real robot) there is no lack of generality.

Interactions among the teammates are made

through the action bounding regions, that is each ro-

bot has access to the bounding region being used by

any teammate.

The robots use a single action, denoted a

◦

, deﬁned

◦



F (q)=(G − q) ∩ H(q)

B(q)={p|p = q + αG(q),α∈ [0, 1]}

(10)

where q is the conﬁguration of the robot, G stands for

a goal set, and H(q) stands for the set of admissible

motion directions at conﬁguration q (easily obtained

from the well known kinematics model of the uni-

cycle). This action simply yields a motion direction

pointing straight to the goal set from the current robot

conﬁguration. Whenever there are no admissible con-

trols, i.e., the set of admissible motion directions that

lead straight to the goal region is empty, F (q)=∅,

the bounding region is expanded using the action



(q) ∪

set of motion

directions

| d(H

,G− q) → 0

if q  B

(q) otherwise

where d(, ) stands for a distance between the sets in

the arguments. This action corresponds to having

H(q) converging to G − q. Generic algorithms al-

ready described in the literature can be used obtain

this sort of convergence. The same set of actions and

the supervisor maps is used by all the robots.

In the simulation experiments robots have access to

a synthetic image representing a top view of the en-

vironment and start their mission at the lower part of

the image. The irregular shape in the upper part of

the images shown in the following sections represents

the raw mission goal region to be reached by the ro-

bots. Basic image processing techniques are used to

extract a circle goal region, centered at the centroid

of the convex hull of the countour of these shapes.

This circle is shown in light colour superimposed on

the corresponding shape. The simulations were im-

plemented as a multi-thread system. Each robot sim-

ulator thread runs at an average 100 Hz whereas the

architecture thread runs at an average 80 Hz. Exper-

iments data is recorded by an independent thread at

100 Hz.

ICINCO 2006 - ROBOTICS AND AUTOMATION

In the real experiment, the image data is used only

to extract a goal region for the robot to reach. The

navigation relies on odometry data.

4.1 Mission 1

The purpose of this experiment is to assess the behav-

ior of the team when each robot operates isolatedly.

Each robot tries to reach the same goal region (in the

upper righthand side of the image). No information

is exchanged between the teammates and no obstacle

avoidance behaviors are considered. In the simulation

experiments robots 1 and 2 start with a 0 rad orienta-

tion whereas robot 0 starts with π rad orientation.

Figure 1 shows the trajectories generated by the ro-

bots. The symbol ◦ marks the position of each robot

along the mission. Dashed lines connect the time re-

lated robot position marks.

The supervisor maps were chosen as,

◦

≡ a

◦

if H(q) ∩ G − q = ∅



≡ a



if H(q) ∩ G − q = ∅

(11)

The oscilations in all the trajectories in Figure 1

result from the algorithm used to expand the action

bounding region at the initial time as the initial con-

ﬁguration of any of the robots does not point straight

to the goal set.

Robots trajectories

50 100 150 200 250 300 350 400 450 500

100

150

200

250

300

350

400

450

500

robot 2

robot 1

robot 0

goal region

Figure 1: Independent robots.

4.2 Mission 2

The robots have no physical dimensions but close po-

sitions are to be avoided during the travel. This con-

dition is relaxed as soon as the robot reaches the goal

set, that is, once a robot reaches the goal region and

stops the others will continue trying to reach the goal

independently of the distance between them.

The supervisor maps are identical to the previous

mission. However, the co-domain of s

◦

, the set A

◦

is changed to account for the interactions among the

robots.

◦





− q



∩ H

)

)\∪

j=i

)

(12)

where the superscript indicates the robot the action

belongs to.

Actions in (12) use a subset of the motion direc-

tions in (10). The bounding region is constructed

from that in (10) including the information from

the motion trend of the teammates by removing any

points belonging also to that of the teammates. The

original mission goal is replaced by intermediate goal

regions, G

, placed inside the new bounding region.

As the robots progress towards these intermediate

goals they also approach the original mission goal. A

potential drawback of this simple strategy is that the

smaller bounding region constrains signiﬁcatively the

trajectories generated.

Figure 2 illustrates the behavior of the team when

the robots interact using the actions (12). The global

pattern of the trajectories shown is that of a line for-

mation. Once the robots approach the goal region the

distance between them is allowed to decrease so that

each of them can reach the goal.

Robots trajectories

50 100 150 200 250 300 350 400 450 500

100

150

200

250

300

350

400

450

500

robot 2

robot 1

robot 0

goal region

Figure 2: Interacting robots.

Figure 3 illustrates the behavior of the team when

a static obstacle is present in the environment. Sim-

ilarly to common path planning schemes used in ro-

botics, an intermediate goal region is chosen far from

the obstacle such that the new action bounding region

allows the robots to move around the obstacle.

From the beginning of the mission robots 0 and 1

ﬁnd the obstacle in their way to the goal region and

select the intermediate goal shown in the ﬁgure. Af-

ter the initial stage where the robots use the expansion

SEMIOTICS AND HUMAN-ROBOT INTERACTION

action aiming at aligning their trajectories with the di-

rection of the goal. The interaction that results from

the exchange of bounding regions is visible similarly

as in the previous experiment. Robot 2 is not con-

strained by the obstacle and proceeds straight to the

goal. Its inﬂuence in the trajectories of the teammates

around the intermediate goal region is visible in the

long dash lines connecting its position with those of

the teammates.

Robots trajectories

50 100 150 200 250 300 350 400 450 500

100

150

200

250

300

350

400

450

500

robot 2

robot 1

robot 0

obstacle

goal region

intermediate

goal region

Figure 3: Interacting robots in the presence of a static ob-

stacle.

4.3 Mission 3

This experiment, with a single robot, aims at validat-

ing the assessment of the framework done, in simula-

tion, in Sections 4.1 and 4.2. The supervisor maps are

those in (11).

Figure 4 shows the Scout robot and the goal re-

gion as extracted using basic color segmentation pro-

cedures. No obstacles were considered in this experi-

ment.

The corresponding trajectory in the workspace is

shown in Figure 5. The robot starts the mission with

orientation π/2, facing the goal region, and hence

maneuvering is not required. The robot proceeds di-

rectly towards the goal region. Figure 6 displays the

trajectory when the robot is made to wander around

the environment. Goal regions are deﬁned at random

instants and when detected by the robot its behavior

changes to the goal region pursuing. The plot shows

the positions of the robot when the targets were de-

tected and when they are reached (within a 5cm error

distance). The visual quality of these trajectories is

close to those obtained in simulation which in a sense

validates the simulations.

Figure 4: A goal region deﬁned over an object in a labora-

tory scenario.

0.5 1 1.5 2 2.5 3 3.

y (m)

x (m)

SIB_log5 trajectory

starting region

goal region

Figure 5: Moving towards a goal region.

0 1 2 3 4 5 6

SIB_log_full trajectory

y (m)

x (m)

target 1

detected

target 2

detected

target 1

reached

target 2

reached

Figure 6: Moving betwen multiple goals regions.

ICINCO 2006 - ROBOTICS AND AUTOMATION

5 CONCLUSIONS

The paper presented an algebraic framework to model

HRI supported on semiotics principles. The frame-

work handles in a uniﬁed way any interaction related

to locomotion between a robot and its external envi-

ronment.

The resulting architecture (see Figure 9) has close

connections with multiple other proposals in the liter-

ature, namely in that a low level control layer and a

supervision layer can be easily identiﬁed.

Although the initial conﬁgurations do not promote

straight line motion, the simulation experiments pre-

sented show trajectories without any harsh maneuver-

ing. The experiment with the real robot validates the

simulation results in the sense that in both situations

identical actions are used and the trajectories obtained

show similar visual quality.

Future work includes (i) analytical study of control-

lability properties in the framework of hybrid systems

with the continuous state dynamics given by differ-

ential inclusions, (ii) the study of the intrinsic prop-

erties for the supervisor building block, currently im-

plemented as a ﬁnite state automata, that may sim-

plify design procedures, and (iii) the development of

a basic form of natural language for interaction with

robots given the intuitive meanings that can be given

to the objects in the framework.

ACKNOWLEDGEMENTS

This work was supported by the FCT project

POSI/SRI/40999/2001 - SACOR and Programa Op-

eracional Sociedade de Informac¸

ao (POSI) in the

frame of QCA III.

REFERENCES

Albus, J. (1987). A control system architecture for intelli-

gent systems. In Procs. of the 1987 IEEE Intl. Conf.

on Systems, Man and Cybernetics. Alexandria, VA.

Aylett, R. and Barnes, D. (1998). A multi-robot architec-

ture for planetary rovers. In Procs. of the 5th Eu-

ropean Space Agency Workshop on Advanced Space

Technologies for Robotics & Automation. ESTEC,

The Netherlands.

Chandler, D. (2003). Semiotics, The basics. Rutledge.

Codognet, P. (2002). The semiotics of the web. In

Leonardo, volume 35(1). The MIT Press.

Goguen, J. (2003). Semiotic morphisms, representations

and blending for interface design. In Procs of the

AMAST Workshop on Algebraic Methods in Language

Processing. Verona, Italy, August 25-27.

Hager, G. and Peterson, J. (1999). Frob: A transformational

approach to the design of robot software. In Procs.

of the 9th Int. Symp. of Robotics Research, ISRR’99.

Snowbird, Utah, USA, October 9-12.

Huntsberger, T., Pirjanian, P., Trebi-Ollennu, A., Nayar, H.,

Aghazarian, H., Ganino, A., Garrett, M., Joshi, S., and

Schenker, P. (2003). Campout: A control architecture

for tightly coupled coordination of multi-robot sys-

tems for planetary surface exploration. IEEE Trans.

Systems, Man & Cybernetics, Part A: Systems and Hu-

mans, 33(5):550–559. Special Issue on Collective In-

telligence.

Kortenkamp, D., Burridge, R., Bonasso, R., Schrekenghost,

D., and Hudson, M. (1999). An intelligent software

architecture for semi-autonomous robot control. In

Procs. 3rd Int. Conf. on Autonomous Agents - Work-

shop on Autonomy Control Software.

Makatchev, M. and Tso, S. (2000). Human-robot interface

using agents communicating in an xml-based markup

language. In Procs. of the IEEE Int. Workshop on Ro-

bot and Human Interactive Communication. Osaka,

Japan, September 27-29.

Malcolm, G. and Goguen, J. (1998). Signs and representa-

tions: Semiotics for user interface design. In Procs.

Workshop in Computing, pages 163–172. Springer.

Liverpool, UK.

Meystel, A. and Albus, J. (2002). Intelligent Systems: Ar-

chitecture, Design, and Control. Wiley Series on In-

telligent Systems. J. Wiley and Sons.

Neum

uller, M. (2000). Applying computer semiotics to hy-

pertext theory and the world wide web. In Reich, S.

and Anderson, K., editors, Procs. of the 6th Int. Work-

shop and the 6th Int. Workshop on Open Hypertext

Systems and Structural Computing, Lecture Notes in

Computer Science, pages 57–65. Springer-Verlag.

Nicolescu, M. and Matari

c, M. (2001). Learning and inter-

acting in human-robot domains. IEEE Transactions

on Systems, Man, and Cybernetics, Part A: Systems

and Humans, 31(5):419–430.

Peterson, J., Hudak, P., and Elliott, C. (1999). Lambda in

motion: Controlling robots with haskell. In First In-

ternational Workshop on Practical Aspects of Declar-

ative Languages (PADL).

Smirnov, G. (2002). Introduction to the Theory of Differ-

ential Inclusions, volume 41 of Graduate Studies in

Mathematics. American Mathematical Society.

SEMIOTICS AND HUMAN-ROBOT INTERACTION