A Multiagent System to Model Human Action Based on the Concept of
Affordance
Zoubida Afoutni
1
, R
´
emy Courdier
1
and Franc¸ois Guerrin
2
1
Universit
´
e de la R
´
eunion, Laboratoire d’Informatique et de Math
´
ematique,France
2
Centre de coop
´
eration internationale en recherche agronomique pour le d
´
eveloppement (CIRAD),
UR Recyclage et risque, Montpellier, France
Keywords:
Situated Action, Affordance, Emergence, Multiagent System.
Abstract:
This paper presents a model to represent situated action in farming systems. The basic idea is to invert the
classical vision of action that stipulates that action is a consequence of a decision process and a plan. We
consider action as mainly dictated by a situation created by the interaction of the actor and his/her environment.
Hence, action is a process that emerges from this situation. Thus, in this paper we treat the question of how to
model action as an emergence process from the situations created by actors and their environment? For this,
our model is based on multiagent system as well as on the affordance, emergence and stigmergy concepts (this
paper emphasizes mainly the first two).
1 INTRODUCTION
Human action is the driving force of many systems,
such as farming systems. Understanding these sys-
tems, namely with the aim of providing management
support, requires a representation of human action as
close to reality as possible. Thus we propose here a
model to represent human action, taking into account
explicitly its spatial and temporal dimensions. Our
conception of human action is based on the theory of
situated action (Suchman, 1987) which assumes that
a realistic representation of action should not extract
it from its occurrence situation nor reduce it to a pre-
constructed plan. A defining feature of this theory is
the concept that all resources required for action are
located in the actor’s environment. The latter is able
to perform the appropriate action at any time, through
the observation of his/her local environment. Thus it
is the actor’s environment that, in some way, dictates
action, while the actor is performing it. This work is
based precisely on the idea that the choice of the ac-
tion to be performed, interrupted or cancelled is not
the responsibility of the actor but rather of its envi-
ronment. To implement our model we used a multi-
agent systems (MAS) modelling approach, along with
the concepts of affordance, emergence and stigmergy.
The first two concepts were used to represent action at
an individual level. The concept of stigmergy is used
to represent action at a collective level, specifically
the indirect coordination between actors mediated by
marks left in the environment. As this paper is fo-
cused on the representation of action at the individual
level, therefore the coordination mechanisms will not
be discussed (Afoutni, 2012; Afoutni et al., 2010).
The choice of MAS was natural because it allows one
to implement explicitly everything related to human
activity and, as we will demonstrate, to situated action
as well. Indeed, an action is performed by an actor
embodied within a space. Its location limits his/her
perception and therefore all actions that he/she can
achieve. The MAS approach allows us to design our
model using multiagent basic concepts: Agent, En-
vironment and Interaction. Moreover, since we con-
sider the actor’s environment plays a crucial role in
determining action in time and space, the environment
must embody all the information needed for the ac-
tor to perform the appropriate actions. The concept
of affordance (Gibson, 1986) is an elegant solution to
implement our idea. Indeed, according to Gibson, an
affordance represents a set of actions offered to an ac-
tor by the objects in his/her environment at a given
time and location. This concept has been developed
in various works to determine that affordance is rather
an emergent property of the pair actor-environment
(Stoffregen, 2003; Chemero, 2003). The concept of
affordance is closely related to the notion of situa-
tion defined as a set of resources and constraints with
which an actor interacts (Afoutni, 2012).
644
Afoutni Z., Courdier R. and Guerrin F..
A Multiagent System to Model Human Action Based on the Concept of Affordance.
DOI: 10.5220/0005141606440651
In Proceedings of the 4th International Conference on Simulation and Modeling Methodologies, Technologies and Applications (SIMULTECH-2014),
pages 644-651
ISBN: 978-989-758-038-3
Copyright
c
2014 SCITEPRESS (Science and Technology Publications, Lda.)
Since the objects in an actor’s environment are an
integral part of the situation he/she encounters, defin-
ing a mechanism that could detect these affordances
will lead to representing situated action. Wishing to
answer the question: how can we detect affordances
in the actor-environment system, we focused on the
concept of emergence which is a promising way to
model affordances. This paper is organized in three
sections. The first section introduces briefly the con-
cept of affordance followed by the relationship be-
tween affordance and emergence. The second section
presents our model of representation of situated ac-
tion. The third section is dedicated to the experimen-
tation of the simulation model, and we conclude this
work by a discussion.
2 AFFORDANCE AND
EMERGENCE
The concept of affordance has its roots in the work
of the psychologist Koffka (Koffka, 1935). However,
Gibson improved this concept for his work on human
direct visual perception (Gibson, 1986). The hypoth-
esis of direct perception outlines that there is no need
of a symbolic representation of the environment in
the actor’s mind nor an inference mechanism to ex-
tract the meaning of his/her percepts. Instead, any
solicitation from the environment is perceived as an
affordance directly by the actor. For example, a rigid
horizontal surface affords an actor to walk; a chair af-
fords a person to sit while it affords an ant to climb.
However there is no consensus on ” where is exactly
the affordance in the pair actor-environment ”, al-
though there is unanimity that affordance is an emer-
gent property of the actor-environment system (Luyat
and Regia-Corte, 2009; Stoffregen, 2003). Accord-
ing to these authors, affordances exist only in the pair
actor-environment, as the fruit of their interactions.
This suggests that emergence can only be observed by
a higher level than the actor-environment interaction’s
level. An affordance is thus not reducible to only one
element of the actor-environment pair. We find simi-
larities between this notion of affordance and the con-
cept of emergence. The latter states that a system ex-
hibits emergence when there are phenomena (proper-
ties, structures, behaviours, etc.) that arise from the
interactions between the entities of the system at the
”micro” level and become visible only at a ”macro”
(higher) level. Emerging phenomena are character-
ized by their novelty compared to the system’s indi-
vidual entities (Wolf and Holvoet, 2005).
Thus, since an affordance is a property originat-
ing from the pair actor-environment and since it is re-
lated to an actor, the latter is capable of detecting this
affordance and exploits it. Hence, the actor plays a
double role. It is an element of the system leading
to the emergence of the affordance as well as the ob-
server who is able to discover this affordance through
its knowledge on the system, as the work of (David
and Courdier, 2009) suggests. These authors present
the idea that experts are sometimes able to discover
or observe an emergent phenomenon in some areas.
This discovery would not have been possibly done
by other people who do not have the necessary meta-
knowledge of the expert’s domain. David et al. pro-
posed to define emergence as a meta-knowledge and
a methodology for detecting and reifying emerging
phenomena on which we have based our work. Our
model exploits all these concepts. However, accord-
ing to our conception, an action is dictated by the en-
vironment and not by the actor. Thus we consider that
the meta-knowledge which allows one to discover an
affordance must be specified relative to the spatial en-
vironment.
3 THE MODEL
Our model is based on the idea that a multi-agent
system has an environment composed of a set of
non-autonomous entities called environmental enti-
ties, representing actors and passive entities of the
studied system. The environmental entities are lo-
cated in a physical space in which they interact. This
environment is controlled by a set of abstract agents
called place-agents. Each place-agent is responsible
for the entities operating in a well-defined area that
we call place. The place-agent’s role is:
• to observe within its perception field (its own
place and neighbourhood) the environmental en-
tities and their interactions in order to detect and
reify the possibly emerging affordances as well as
to select and trigger the appropriate actions on the
place it manages;
• to coordinate with other place-agents hence giv-
ing indirectly a consistent behaviour to the entire
system.
This model therefore proposes to reverse the tra-
ditional view of MAS where the real-world actors
are usually represented by ”agents” (in the computer-
science meaning) whereas the other system entities
are represented as ”objects of the environment”. For
us, the real-world actors are represented by non-
autonomous entities the role of which is to perform
the actions dictated by the place-agents.
AMultiagentSystemtoModelHumanActionBasedontheConceptofAffordance
645
We present the model in four parts: environment,
affordance, representation of action and agent archi-
tecture.
3.1 Environment
The environment is composed of environmental enti-
ties as well as the physical space where they are lo-
cated.
3.1.1 Environmental Entities
Environmental entities represent the ”actors” and all
kind of passive objects of the studied system. An
environmental entity is characterized by a dynamic
state and processes representing their internal dynam-
ics. We distinguish two environmental entity types:
passive-object and actuator. A passive-object repre-
sents an entity that can only undergo an action. In
the case of agricultural systems, a passive-object can
be a stock of fertilizer, a plot, a tractor, etc. An ac-
tuator is an environmental entity which possesses all
mechanisms to implement an action but not the ini-
tiative capacity to undertake it which is the place-
agents’ role. Thus, an actuator may be a combina-
tion of an actuator with one or more passive-objects.
For example, the action of ploughing can be done by
the actuator ”farmer-tractor-plough”. If ”farmer” is
an actuator per se, the particular action of ploughing
needs to combine it with ”tractor” and ”plough” other-
wise considered as passive-objects. This environmen-
tal entity characterization allows one to differentiate
between entities that perform action from entities that
undergo action. This helps answer a question related
to action ”who performs action? ” The answer is the
pair actor-environment likely to making affordances
at some time and some place to emerge. Therefore,
environmental entities differ from agents by their lack
of ability to take the initiative of acting (for actuators)
and even to act (for passive objects).
Definition 1. An environmental entity is a situated
entity that carries information useful for the emer-
gence of affordances. It is capable to execute or
undergo an action, with internal processes that can
modify its environment. It is characterized by the lack
of autonomy and faculty of perception.
An environmental entity e is characterized by:
• id
e
an unique identifier represented as an al-
phanumeric chain;
• A
e
a set of actions that it can perform or undergo:
A
e
= {a
e
1
, a
e
2
, ..., a
e
n
};
• E
e
a set of variables that model its state:
E
e
= {e
f
1
, e
e
2
, ..., e
e
n
};
• D
e
a set of functions that calculate the evolution
of its internal state: D
e
: E
e
→ 2
E
e
;
• I
f
a set of functions that connect one of its state at
a time t with the list of actions it can perform or
undergo:
I
f
: E
e
× t → A
e
• p
e
(t) a place where it is located at time t.
3.1.2 Spatial Representation
Representation of situated action necessarily involves
the explicit representation of space. Thus, the phys-
ical space is where the environmental entities are lo-
cated. For example in agriculture, space is the set of
land plots, buildings, equipment, roads, streams, etc.
that compose the farm. The space in the model is a
two dimensional Euclidean space. Each environmen-
tal entity is characterized by its position defined in a
coordinate system. Space is regarded as continuous
even if the environment is managed by place-agents.
Each place-agent controls only a portion of space (its
place) characterized by a vector of coordinates (see
figure 1). Each place can have any geometric shape.
Therefore the space composed by a set of places of
equal sizes, with the same geometric shapes, is a dis-
crete space (as mostly used in MAS); otherwise it is
an irregular space as our model gives the modeller this
possibility (see figure1).
Figure 1: Partition of continuous space into places.
Definition 2. A place p is defined by :
• id
p
an identifier of the place;
• v
p
a vector of coordinate points that specifies the
geometry of p;
SIMULTECH2014-4thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
646
• id
g
an identifier of the agent managing the place
p;
The space P is a set made of the union of all ele-
mentary places p
i
.
3.2 The affordance Model
Definition 3. An affordance is an emergent property,
which can be detected at a macro level using meta-
knowledge held by agents-places that control the por-
tions of the physical space (called ”places”) where
environmental entities are located.
Modelling affordances as emergent phenomena
leads us to adapt the mechanisms involved in the de-
tection and reification of emergent phenomena(David
and Courdier, 2009).
3.2.1 Detection of Affordance
The detection of affordances by place-agents consists
in observing at all time the states of all the environ-
mental entities located on the place they manage or
their immediate neighbourhood making the list of ac-
tions that can be performed or dealt with by them.
This observation allows place-agents to build a set
of facts K that represent the knowledge embedded in
their environment and to discover affordances.
A fact is characterized by the 3-tuple (environ-
mental entity, state/action).
After building K, we need to analyze this set to
discover potential affordances. We use a set of func-
tions R
Φ
called affordance revealing functions. Every
function’s argument is the set K of facts and its output
is a set of potential affordances.
R
Φ
=
{
f : K
n
→ Φ
}
, n ∈ N (1)
Where Φ is the set of emerging affordances.
3.2.2 Reification of Affordance
The affordance detection phase outlines new affor-
dances that will be used by the agent:
• to select one or more actions to perform;
• to improve the system with new knowledge that
did not exist originally.
The second point is the key advantage of reification.
Detected affordances will be reified into the environ-
ment through what we call a structure of affordance,
characterized by: (i) the affordance (i.e. a possible ac-
tion) (ii) the actuator(s)/passive object(s) causing the
detected affordance.
Reification of affordances is subject to a set of
conditions that specify whether an affordance can be
reified and when it should be removed from the en-
vironment. Indeed, a structure of affordance may be-
come invalid when the current state does not match
the conditions that create this affordance.
Reification allows us to create knowledge in the sys-
tem. The structures of affordances allow the agent
to discover thereafter other affordances that could not
have been detected with the basic entities (actuators
and passive objects) of the system. The structures of
affordances also mimic a memory embedded in the
environment that can be used directly by an agent
when the configuration of the environmental entities
located on its place has not changed between two sim-
ulation time steps.
3.3 Representation of Action
In this paper, we consider action as a non-
instantaneous dynamic process, whose execution is
constrained by the perceived conditions in the agent’s
environment. Thus, an action has a start date and end
date and a duration. It may be interrupted or can-
celled if the conditions necessary for its continuation
no longer exist. An action may be in different states:
running, interrupted, canceled.
Definition 4. An action is a dynamic process invoked
by an agent, executed by an actuator, which can be
undergo by a passive-object, characterized by a start-
ing date, an ending date and a duration, conducted in
a place of the space. It possesses a state and a prior-
ity.
3.3.1 Dynamic State of an Action
Our model is based on the representation formalism
of action presented in (Guerrin, 2009). An action a is
represented by a binary function of time (t), space (P)
an environmental entity (E) and constrained by a set
of conditions C
a
(t, p, e):
S
a
(t, p, e) =
1
0
if
else
C
a
(t, p, e)
(2)
p is the place where action is running, t the current
time, e the set of environmental entities involved in
the action. C
a
(t, p, e)) is a logical proposition depend-
ing on time and space. When this proposition equals
1, it means that the current situation allows the exe-
cution of action. Otherwise, three cases are possible:
(i) the current state no longer allows the continuation
of action (e.g. the current date value is greater than
the end date), then the action will be definitively can-
celled (ii) the current situation corresponds to the sit-
uation for the termination of an action, then the end
AMultiagentSystemtoModelHumanActionBasedontheConceptofAffordance
647
date of action will be calculated (iii) the current sit-
uation requires the interruption of action. Thus, an
action a is represented as a dynamic process defined
by a succession of time intervals.
3.3.2 Action Temporal Bounds
The starting and ending times (t
−
,t
+
) and the duration
(τ
a
(t)) of an action are also functions of time, deter-
mined by a condition P
±
a
.
t
±
a
(t, p, e) =
n
t if P
±
a
(t, p,e)
t
±
a
(max(0, t−τ
a
)) else
(3)
When the condition P
−
a
(t, p, e) (resp P
+
a
(t, p, e))
is satisfied it can trigger the corresponding action and
calculate its start (or end) date, otherwise the value t
±
a
is identical to its value at the previous time step.
3.4 Agent Architecture
Physical space and environmental entities are man-
aged by a set of place-agents. Each of them is re-
sponsible for managing actions on its assigned place
(figure 2). These actions are induced by the environ-
mental entities and other conditions observed on its
place through detected affordances. Thus, the detec-
tion of affordances, the selection and update of the
state of action as well as its place’s state are the main
duties of any place-agent.
3.4.1 Perception
When a place-agent perceives an environmental en-
tity, it may arise two cases:
• the environmental entity e perceived is on the
place managed by the place-agent g;
• the environmental entity e perceived is on the
place managed by a neighbouring place-agent g
0
.
In the first case, it could lead to trigger an action
by the actuator bearing the capacity of performing it.
In the second case, since a place-agent does not have
the right to trigger an action with an actuator that is
not on its own place, this could lead to initiate a be-
haviour of coordination with the neighbouring agent.
Hence, we define two boolean perception functions:
Perc
e
and Obs
e
.
Perc
e
(g, e) =
1 if e on place of g
0 else
(4)
Obs
e
(g, e) =
1 if e on neighbouring place of g
0 else
(5)
Once the environmental entities that are perceived
by a place-agent are identified, the place-agent needs
to collect facts potentially leading to the emergence of
an affordance. A place-agent can thus perceive only
the facts necessary to allow the emergence of affor-
dances corresponding to the actions that could be per-
formed on passive-objects or performed by actuators
located on this place. The facts perceived by a place-
agent should be restricted by a set of filters Filtre
K
.
These are functions that specify for each agent the
facts at the environmental entity’s level that it can per-
ceive. These filters are defined by:
Filter
K
: G × E → K (6)
3.4.2 Detection of Affordances
The perception phase allows a place-agent to acquire
information on its local environment which represents
all the facts. These facts may concern the informa-
tion collected at environmental entities’ level or about
the general state of the environment. The facts from
the environmental entities will allow the agent to help
affordances to emerge using the affordance revealing
functions R
φ
.
3.4.3 Selecting an Action
Once the list of affordances has been built, the place-
agent needs to select one or more actions which can
run concurrently. This selection is based on both pre-
conditions necessary for triggering actions P
±
a
and the
priorities assigned to actions.
When priorities are assigned a priori they are
fixed and do not change during the simulation. But
they could be dynamically assigned based on cer-
tain time-varying parameters. For example, some ac-
tions in the agricultural system cannot be performed
outside of their earliest and latest completion dates.
These actions become urgent when their end dates at
the latest approach. The priority of an action is known
using a function Priorite
a
that returns an integer.
The end of the selection phase of action is marked
by the calculation of its starting date, using the equa-
tion (3).
3.4.4 Triggering an Action
After selecting an action and calculating its starting
date, the action will be triggered at the concerned ac-
tuator’s level. Concretely triggering the action con-
sists in bringing to ’1’ its state and in triggering
processes implemented at the environmental entities’
level representing the effects of this action on the en-
vironment.
SIMULTECH2014-4thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
648
Figure 2: The relationship between the layer of agent and the environment.
3.4.5 Checking During an Action
Given that action depends on a situation, its course
is influenced by the evolution of this situation. Thus,
the place-agent may be required to suspend, cancel or
terminate the current action. This phase consists in
checking at the beginning of each simulation cycle if
the conditions of action’s execution are still valid. If
they are no longer valid, there are two cases:
• the place-agent interrupts its action;
• the place-agent notes that the conditions for com-
pletion of action P
+
a
(t, p, e) are valid. It calculates
the ending date using equation (3) and informs the
environmental entity concerned of the end of ac-
tion.
4 CASE STUDY
4.1 Implementation
A prototype of the proposed model has been im-
plemented using the simulation platform AnyLogic
(AnyLogic, 2010). AnyLogic can simulate models
based on multiple modelling approaches: discrete
event simulation as well as systems dynamics and
multi-agent systems. It is based on the object-oriented
language Java. Regarding the multi-agent approach,
AnyLogic offers two types of entities: Agent and Ac-
tiveObject. ActiveObject can be understood as ob-
jects in object-oriented programming. The behaviour
of objects (i.e. any entity of the model) can be imple-
mented using state charts, simple functions, data flow
diagrams, activity diagrams, etc.
4.2 Description of the Simulated System
Figure 3 gives an overview of the simulation of a
model based on a GIS map with two farms. Farm
1 is made of 4 plots and 1 livestock building. Farm 2
is made of 3 plots.
The simulation aim is to represent the actions ex-
ecuted at the farm level concerning the livestock (e.g.
emptying manure stocks) and crop cultivation of car-
rots and potatoes (e.g. ploughing, fertilization, etc.).
4.3 Modelling of the Simulated System
4.3.1 Environment
The environment is made by the physical 2-D space
as well as all environmental entities. In this model
implementation, the physical space is actually par-
titioned into regular square-shaped cells although
it is considered as continuous. Passive-objects are
the stocks of manure, of fertilizers, the crop plots,
etc. Actuators are the entities that can perform
actions such as the farmer, the farmer and his tractor
equipped with plough or other tools, etc.
4.3.2 Agents
We have three types of agents: plot-agent, building-
agent and road-agent. The distinction between these
three types is based on the actions that can be per-
formed on the place controlled by every type of
agents. For example, plot-agents have the ability
to ”fertilize the soil” but not to ”empty the manure
stock” (in fact to order relevant actuators to do that
for them). But the latter action can be ordered by the
livestock-building-agent. Thus, the type of agent con-
strains its perceptions and therefore all the facts K that
can be constructed at each simulation time step.
The affordance revealing and reification functions
are based on the type of agent. Consequently, emerg-
ing affordances on a place p correspond to the actions
that can be performed on p. Each agent is assigned a
place to manage, as shown in figure 3 on the cell grid
representing space in this example.
AMultiagentSystemtoModelHumanActionBasedontheConceptofAffordance
649
Figure 3: Simulation interface of the prototype implementing our action conceptual model using the AnyLogic platform.
4.4 Output
The graph 1 in figure 3 shows the evolution of the
manure and livestock feed stocks that are located at
the places managed by the livestock-building-agent.
The behaviour of these two entities was implemented
using the data flow diagrams shown on figure 3.
Graph 2 shows the actions running in farm 1 (soil
disinfection).
Graph 3 shows the actions running in Farm 2
(ploughing).
These graphs illustrate the actions performed in
the agricultural system simulated. They show that the
implemented MAS based on our conceptual model
can simulate in a consistent manner a set of two farm-
ing systems without using any planning process. Al-
though the capability of the model to account for the
coordination of actions (namely implicit coordination
based on stigmergy) has not yet been demonstrated
here, this first result is a preliminary proof-of-concept
of our model.
5 DISCUSSION
Our paper focuses on situated multi-agent systems
(Weyns et al., 2007). One key-aspect that character-
izes situated MAS is the role of the environment. Sev-
eral functions are assigned to it: structuring entities
within the system, communication and coordination
between agents, etc.
We have shown that using the concepts of affor-
dance and emergence allows us to create a ’smarter’
environment for the agents using the knowledge car-
ried by actuators and passive objects. The affor-
dance detection mechanisms allow emerging actions
of agents to be performed in time and space. Reifi-
cation of these affordances as ’structures of affor-
dances’ provides the agents with a memory of past
affordances. This avoids agents to recalculate their
affordances at the next simulation step when nothing
has changed in the meanwhile. Thus, it allows the
modeller to add in the simulation scenarios as many
actions as necessary without affecting the design of
agents and the environment represented in the model.
Indeed, we only have to inject new affordance reveal-
ing and reification functions to modify the simulation
scenarios. Therefore a modification of the studied
system does not imply its global destabilization nor
restarting its whole design.
One model feature lies in the fact that the decision
of the action to be performed is attributed to the en-
vironment through the concept of place-agents. Each
place-agent manages a part of the physical space spec-
SIMULTECH2014-4thInternationalConferenceonSimulationandModelingMethodologies,Technologiesand
Applications
650
ified by a place where environmental entities may be
located be they mobile or static. This is in perfect har-
mony with the theory of situated action which states
that actions are mainly determined at each instant by
the situation in which the actor is. Hence, no overall
strategy or central plan is necessary. Several other ad-
vantages can be dragged out from such a representa-
tion: the ability to simulate complex activity systems,
to process both parallel and sequential actions and in-
teractions, to account for interruption and resumption
of actions, delays, etc.
Our prototype allowed us todate to validate the
proposed concepts on a simple case study. To make
it more useful to agronomists, the next step is to val-
idate the model on more realistic complex situations.
Moreover, the proposed conceptual model has been
designed on a level of abstraction independent of the
field of agricultural applications. We also plan to test
our approach in other areas involving human actions.
ACKNOWLEDGEMENTS
The authors thank Dr. Yassine Gangat that helped sig-
nificantly to improve the English text of this article.
REFERENCES
Afoutni, Z. (2012). Modelling situated action based on af-
fordances and stigmergy. In IEEE Sixth International
Conference on Self-Adaptive and Self-Organizing Sys-
tems, pages 175–180.
Afoutni, Z., Martin-Clouaire, R., Courdier, R., and Guerrin,
F. (2010). Coordination of activities: application of
some concepts and formalizations to agricultural sys-
tems simulation. In International Conference on Mod-
eling and Applied Simulation, pages 25–31.
AnyLogic (2010). http://www.anylogic.com.
Chemero, A. (2003). An outline of a theory of affordances.
Ecological Psychology, 15(2):181–195.
David, D. and Courdier, R. (2009). See emergence as a
metaknowledge - A way to reify emergent phenomena
in multiagent simulations? In International Confer-
ence on Agents and Artificial Intelligence, pages 564–
569.
Gibson, J. (1986). The Ecological Approach to Visual
Perception, Resources for Ecological Psychology.
Lawrence Erlbaum and Associates, New Jersey.
Guerrin, F. (2009). Dynamic simulation of action at op-
erations level. Autonomous Agents and Multiagents
Systems, 18(1):156–185.
Koffka, K. (1935). Principles of Gestalt psychology. Har-
court, Brace and Company, NewYork.
Luyat, M. and Regia-Corte, T. (2009). Les affordances: de
James Jerome Gibson aux formalisations r
´
ecentes du
concept. Ann
´
ee psychologique, 109(2):297–332.
Stoffregen, T. (2003). Affordances as properties of the
animal environment system. Ecological Psychology,
15(2):115–134.
Suchman, L. (1987). Plans and situated actions: The prob-
lem of human-machine communication. Cambridge
University Press, Cambridge, MA, USA.
Weyns, D., Omicini, A., and Odell, J. (2007). Environment
as a first class abstraction in multiagent systems. Au-
tonomous Agents and Multi-Agent Systems, 14(1):5–
30.
Wolf, T. D. and Holvoet, T. (2005). Emergence versus
self-organisation: Different concepts but promising
when combined. Engineering self-organising systems,
pages 1–15.
AMultiagentSystemtoModelHumanActionBasedontheConceptofAffordance
651
