VICA: A Vicarious Cognitive Architecture Environment Model for

Navigation Among Movable Obstacles

Halim Djerroud and Arab Ali-Ch

erif

Laboratoire Paragraphe, Universit

e Paris VIII, 2 rue de la Libert

e 93526 Saint-Denis, France

Keywords:

Navigation Among Movable Obstacles, Cognitive Architecture, Multi-agent System.

Abstract:

This article presents a new Cognitive Architecture Environment model for Navigation Among Movable Obsta-

cles (NAMO). This model is the result of a novel approach based on the Theory of Mind and more particularly

on the notion of ’vicariance’ as an essential strategy of the robot’s interaction with outside world. The im-

plementation of our model follows the advances in AI and the Cognitive Robotics research area, where a

cognitive architecture environment is represented as a Multi-Agent System (MAS). The MAS representation

offers the robot the ability to produce a representation of its environment as well as the possibility to run all

types of action simulations in order to anticipate the environment’s reactions. The environment state values,

both predictive and real as transcribed during simulation and real action movements, are compared to each

other in order to keep the correct ones and avoid errors. This is a continuous learning and leads to the con-

struction of a safe path of actions into a dynamic environment. The experiment results show the efﬁciency

of our model, offering an intelligent guide to the robot in order to perform tasks among mobile agents, by

avoiding a maximum number of obstacles.

1 INTRODUCTION

In recent years, a lot of research has dealt with the

goal of how making a robot able to efﬁciently maneu-

ver in a crowded space (Renault et al., 2019), such

as a domestic environment, ﬁlled with different types

of agents such as human, robot, or other. We pro-

pose to reach the same goal from a different point of

view: how the environment, which can not be spec-

iﬁed in advance, could offer to the robot the knowl-

edge gained through interactions with it and the other

agents, and their experience of acting in this environ-

ment. The robot has to manage the possible unpre-

dictability of the objects in the environment and its

actual behavior when it executes its plan. Our ap-

proach is based on the theory of mind (Theory of

Mind - ToM), and more particularly the concept of

vicariance as deﬁned by Berthoz (BERTHOZ Alain,

2015). Vicariance is a polysemous term but funda-

mentally the concept is equivalent to the idea of the

potential substitution of one solution or function for

another. According to this idea, it is possible to per-

form the same tasks with different systems, solutions

or behaviours which constitutes the basis for diversity.

In this article we present the concept and imple-

mentation of a vicarious cognitive architecture model

environment for Navigation Among Movable Obsta-

cles (NAMO), named VICA. The contributions are

the following: 1) The robot has the ability to produce

a mental image of its environment; 2) the robot is able

to run simulations in order to anticipate the environ-

ment’s reactions; 3) The robot interacts with the en-

vironment and learns from all the environment state

values during the training states.

In the domain of cognitive architecture and more

speciﬁcally in cognitive robotics (Lemaignan et al.,

2011), most architectures are generic and few of them

can truly manage the complexity of interactions be-

tween objects being in the same environment. Nu-

merous applications (Mueggler et al., 2014) are cur-

rently developed for robots that move in an envi-

ronment with movable obstacles, but most research

papers dealing with this problem (Moghaddam and

Masehian, 2016)(Mirabel and Lamiraux, 2016) do not

always make explicit the underlying cognitive archi-

tecture.

Our work is part of the theory of mind, which

is also a branch of the philosophy of mind. We

are strongly inspired by the work of the physiologist

Alain Berthoz (Berthoz, 2008). He describes the brain

as a predictor and action simulator. The brain’s func-

tion is to anticipate future environmental events and

simulate the adequate movement to fulﬁll a need, this

is the principle of Vicariance. We represent the cogni-

298

Djerroud, H. and Ali-Chérif, A.

VICA: A Vicarious Cognitive Architecture Environment Model for Navigation Among Movable Obstacles.

DOI: 10.5220/0010269602980305

In Proceedings of the 13th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2021) - Volume 2, pages 298-305

ISBN: 978-989-758-484-8

tive architecture environment model as a MAS system

intelligently reinforced by the continuous simulations

experience acquisition.

2 RELATED WORKS

2.1 Navigation Among Movable

Obstacles (NAMO)

Current work in the domain of NAMO can be divided

into two main categories: ofﬂine planning and online

planning.

Ofﬂine planning assumes that all information

about the space, in which the robot will move, is

known in advance. Among the research on ofﬂine

NAMO, Chen et al. (Chen and Hwang, 1990) ﬁrst

proposed a grid-based method to represent the envi-

ronment, using both a global and local planner, in or-

der to ﬁnd a path that leads to a goal. The algorithm is

not optimal because it does not solve a wide variety of

problems. The work of (Okada et al., 2004) presents

a humanoid robot with three sub-groups and plan-

ners, one for each of the following elements: envi-

ronment, movement and manipulation. The algorithm

does not take into account mobile obstacles that indi-

rectly block the path to the goal(Stilman and Kuffner,

2008). Most results on ofﬂine NAMO research show

that this approach isn’t effective, it lacks ﬂexibility.

In online planning the robot has only partial

knowledge of its environment, and forces the robot

to modify its original plan based on new information

acquired during its journey. The research of Wu et al.

(Wu et al., 2010) is one of the ﬁrst to bring the subject

of NAMO in an unknown environment. It presents

an algorithm based on simple assumptions. The al-

gorithm gathers new information during the robot’s

movement, and identiﬁes new data that do not affect

the calculations of the previously established path.

However the local solutions resulting from new infor-

mation on the immediate environment do not solve

all cases. In similar work for NAMO, in an un-

known environment (Wu et al., 2010), the proposed

algorithm turns out to be optimal under certain con-

ditions, but numerous cases still remain insolvable.

Today’s trend of online NAMO seems more promis-

ing and better-suited. The work of (Levihn et al.,

2013) presents a method providing a theoretical deci-

sion solution for action selection for NAMO applied

to a continuous-control robot. The algorithm com-

bines Markov decision-based planners (MDP) as well

as Monte-Carlo simulation. The presented planners

solve some of the problems encountered in speciﬁc

NAMO cases, but they never present a generic so-

lution applicable in all cases regard of a vicarience

mind.

2.2 Cognitive Architectures

Research in cognitive architectures for NAMO, show

that these approaches offer ﬂexibility in their way to

generalizing problems. Work in the domain of cogni-

tive architecture is longstanding, and state-of-the-art

examples can be found in (Ye et al., 2018)(Kotseruba

et al., 2016). It falls into three main families: 1) Bio-

inspired cognitive architecture. 2) Cognitive architec-

ture for the solving of artiﬁcial intelligence problems.

3) Cognitive architecture based on psychological and

philosophical theories.

2.2.1 Bio-inspired Cognitive Architecture

Has the objective of modelling human behavior, and

has been under continuous development since the late

1970s (Martin et al., 2020) (Remmelzwaal et al.,

2020). Among the best-known, we quote: ACT-R

(Adaptive Control of Thought-Rational) (Anderson,

2019) is organized in a set of modules, each dealing

with a different type of information corresponding to

an equivalent in humans (visual, perception, memory,

manual, etc.). Each module has its own version of

the three memories: Working memory (WM), Declar-

ative memory (DM) and Procedural memory (PM).

Coordination is ensured by a central system. CLAR-

ION (Connectionist Learning with Adaptive Rule In-

duction On-line) (Sun, 2006) is a hybrid architecture

that combines symbolic and connectionist representa-

tions, while developing artiﬁcial agents.

2.2.2 Cognitive Architectures for Solving

Artiﬁcial Intelligence Problems

These architectures are based on logic programming

and machine learning algorithms. We quote SOAR

(State, Operator And Result) (Laird, 2012) which is

a purely ”symbolic AI” architecture, as well as iCub

(Vernon et al., 2011). ICARUS (Choi and Langley,

2018) is more recent, storing two distinct forms of

concepts. They both imply relations between the ob-

jects and need hierarchical organization of long-term

memory. To the best of our knowledge, there is no im-

plementation of these architectures in robots that act

in real environments.

VICA: A Vicarious Cognitive Architecture Environment Model for Navigation Among Movable Obstacles

299

2.2.3 Cognitive Architectures based on

Philosophical Theory

They are based on philosophical and psychological

theories and deal with problems such as action, per-

ception, reasoning and intentionality. In BDI archi-

tecture (Belief, Desire, Intention) (Rao and Georgeff,

1991) we ﬁnd the theory where beliefs and desires are

the cause of the intention to act, like in PRS (Proce-

dural Reasoning System) (Wooldridge, 2009) for ra-

tional agents. LIDA (Learning Intelligent Distribu-

tion Agent) (Friedlander and Franklin, 2008) is a cog-

nitive architecture based on Bernard Baars’s Global

Workspace psychological theory (Baars, 2005). It

has a cognitive cycle divided into three phases: com-

prehension, attention and selection of action, and

learning. These phases are repeated indeﬁnitely.

CARMEL (in French - Compr

ehension Automa-

tique de R

ecits, Apprentissage et Mod

elisation des

Echanges Langagiers) is an architecture developed by

Grard Sabah (Sabah and Briffault, 1993). In this sys-

tem the agent makes itself a symbolic representation

of the one it will interact with.

Currently, researchers also deal with the Theory of

Mind (ToM) which is itself a branch of the philoso-

phy of mind. Our work follows this vein and is in-

spired by the works of physiologists. (Berthoz and

Debru, 2015) describes the brain as a predictor and

action simulator and the main functions are: anticipa-

tion of future events and simulation of the appropriate

movements in order to respond accordingly. The au-

thor calls this principle Vicariance.

3 PRELIMINARY

Neuroscience has inspired many researches in

robotics, the goal of which is to achieve efﬁciency like

natural systems such as the brain. We give some ex-

amples like the role of mental simulation of the road

in navigation (Trullier et al., 1997), or how the brain

simulates Newtonian laws and determines the trajec-

tory of objects in space (McIntyre et al., 2001), or how

it simulates the rotation of an object (Wexler et al.,

1998).

(Berthoz, 2017)(Berthoz, 2012)(Berthoz, 2000) pro-

vides a general theory of the brain functioning. In the

principle of Vicariance, the anticipation and simula-

tion of the appropriate movements to fulﬁll a need.

The simulation of imagining a movement (Wexler

et al., 1998), matches to mentally simulating the

body’s movement in the computational space of the

brain. This mechanism is essential for quick move-

ments; the brain takes all or a part of sensory infor-

mation and process it in order to act. The brain sim-

ulates internally possible actions before choosing and

engaging in one, knowing that in many cases it is not

possible to test multiple actions.

Let us consider a robot using such a system, mov-

ing in a crowded space with movable obstacles. The

robot, facing a dynamic obstacle on a path, must avoid

the obstacle while advancing itself. It considers the

movement of the obstacle to avoid collision when

crossing the obstacle’s path. The function of simu-

lation and prediction is fundamental. The robot plans

the action, while planing the movement at the same

time. Therefore, the system selects what is impor-

tant or pertinent sensory information for this move-

ment. In other words, at every phase of movement, the

brain will pre-select sensory input considered as im-

portant. On the other hand, the system is not limited

to select important sensors only, because it can predict

the state in which the objects should be if the move-

ment is accomplished as they should be. In this article

we propose an implementation based on this hypoth-

esis. To represent the environment (a mental image of

the environment) in terms of data structure, the work

of (Djerroud and Cherif, 2019)(Djerroud and Cherif,

2018) show that MAS are perfectly capable of repre-

senting the environment with its lows and rules. We

seek to reproduce observable environments, in MAS

form, able to modify simulation’s parameters, to un-

derstand its functions, and ﬁnally predict the future

state of the system. The cognitive architecture pre-

sented in this article uses MAS, which offers an envi-

ronmental engine with the possibility to create agents

and integrate rules on the ﬂy. MAS is deﬁned as fol-

lows:

MAS =< Agents, Environment,Coupling >

Agents = Agent

, ..., Agent

Agent

=< State

, Input

, Out put

, Process

Environment =< State

, Process

A MAS is composed of a set of agents, an envi-

ronment and the coupling between them. An agent is

deﬁned as a set of states, inputs, outputs and process.

A state is the set of attributes which deﬁne an agent.

Inputs and outputs are sub-sets of states, whose vari-

ables are coupled with the environment. The inputs

and outputs can represent the sensors and performers

of an agent. The process is an autonomous process

executed internally by the agent. The coupling is a

mechanism enabling the linking of agent attributes to

the environment. The environment is deﬁned as fol-

lows:

Environment =< States, Rules, Process >

States =< SharedAttributes, InternalAttributes >

ICAART 2021 - 13th International Conference on Agents and Artiﬁcial Intelligence

300

The environment is considered an active entity. It

has its own process, so it is able to change its state

independently the actions of the agents that evolve

within. It is composed of states, rules and process.

States are the attributes representing the state of the

environment at a given moment. We can distinguish

two types of attributes: shared attributes and internal

attributes. The former is shared with the agents (po-

sition, for example), while the latter is not. Internal

attributes are the internal properties of the environ-

ment, such as size, coefﬁcient of friction in a certain

space, etc.

Rules are the rules that the environment must re-

spect. For example, two agents must not be in the

same place at the same time. Rules are deﬁned as fol-

lows:

rule =< expression ? action1 : action2 >

A rule is represented as an expression. It corre-

sponds to the law that the environment must check.

The actions correspond to the actions to be performed,

depending on whether the rule is respected or not.

4 VICA ARCHITECTURE:

GLOBAL VIEW

VICA architecture (cf. Fig1) works in a loop. Each

cycle allows: 1) to create an internal representation

of the observed environment, 2) to simulate and ob-

serve the result of these actions, 3) to choose a plan of

actions (among simulated plans of action) and apply

it to the environment. 4) to observe the environment

and evaluate its behavior according to the simulation

(learning).

VICA functions in a deliberative loop, this allows

action simulation plans (an ordered set of actions) and

ﬁnally produces only one. VICA has a speciﬁcity that

enables independent modules to run in a loop. Each

cycle of execution of each module ends with a de-

liberation and the result is delivered in the form of a

message to other modules. Each loop takes as input

perceptions from the environment, which improve in-

ternal representation of the environment. During the

deliberative loop, the cognitive part of VICA, called

Abstract Space (AS) reproduces the perceptions of the

environment in a MAS (Djerroud and Cherif, 2019)

and represents each perceived entity in the environ-

ment as an agent. Subsequently, each agent is en-

riched with all knowledge already known on this type

of agent; this knowledge concerns particularly the

possible actions on this agent. In the case of the detec-

tion of a box, the system already knows the possible

actions, for example push. In the case of the detec-

tion of a person, the knowing actions are for example

asking to move. For now, the associated actions are

integrated in a database. We plan in the short term

to integrate a module allowing to deduce automati-

cally possible actions on an object, commonly called

affordance (Sun et al., 2014). If the object is not ref-

erenced, the possible actions are not known. The sys-

tem will search for the actions of the nearest object;

in VICA this aspect is called vicariance.

VICA is composed of four modules. Each module

functions independently and in a loop. Modules com-

municate with each other with ACL messages. The

ﬁrst module Detection and Agentiﬁcation enables the

observation of the environment and the extraction of

useful information, such as detecting an obstacle. The

second module Abstract Space (AS) can be consid-

ered as a physics engine

. It is able to represent en-

tities in the form of agents and apply laws in order

to observe and predict the evolution of entities in the

environment. The sub-module Simulation enables the

construction of a MAS in terms of observation. The

module Evaluate Outcome indicates to MAS the ac-

tion to be simulated, evaluates the results of different

simulations to make a decision, and then applies the

best action. The ﬁnal module Select Action indicates

how to execute a plan of action in the real environ-

ment.

4.1 Principal Modules of VICA

Architecture

Figure 1: Conception scheme of VICA architecture.

4.1.1 Detection and Agentiﬁcation

The role of this module is to collect and merge the in-

formation coming from the sensors of the robot and

to express them as agents in the MAS integrated in

the AS. More precisely, the mobile robot is equipped

with several sensors (RGB camera, depth camera, LI-

physics engine: software providing an approximate

simulation of real physical systems.

VICA: A Vicarious Cognitive Architecture Environment Model for Navigation Among Movable Obstacles

301

DAR, pressure sensor, etc.). Sensor data is analyzed

to obtain as much information as possible about the

observed scene. An image recognition system is ap-

plied to the RGB camera to determine the type of

object and its shape (box, table, robot, human, etc.),

then the depth camera indicates the dimensions of the

detected objects and ﬁnally the LIDAR indicates the

distance between the moving robot and the objects,

as well as between the objects themselves. The sec-

ond role of this module is Agentiﬁcation which goal

is to present data to AS. When an entity is detected, it

collects as much information as possible about that

entity, sending it to the AS via messages using the

ACL language. Before sending the entities detected

in the scene to the AS module, an object authentica-

tion system adds more information about the possible

actions on these entities. Information is stored auto-

matically to a database, some of it is deduced by a

machine learning algorithm (for example the force to

apply to move an object according to its shape and di-

mensions) or by simple calculations (for example the

volume of a geometric object).

4.1.2 Abstract Space

It enables the construction of an internal representa-

tion of the observed environment. It represents the

current state of the environment in a MAS. Each el-

ement of the environment is represented as an agent.

An agent is therefore considered as an internal repre-

sentation of an object. The ﬁrst part of this sub-system

is responsible for creating agents. When an object is

detected, the system represents it as an agent. There-

fore, we can identify two cases, either the detected ob-

ject already exists in the AS or it is new, in this case an

agent is created or updated. The MAS created in AS

represents a static image of the environment at a given

moment. The Simulation MAS module offers a set of

services that allow the module Evaluate outcome to

accomplish some tasks. Among the offered services :

a) Simulation: enables launching the simulation of an

action (chosen by the Select Action module) b) Result

simulation: enables consulting the attributes (results)

after simulation. c) Commit / RollBack: will allow

the Evaluate outcome module to launch multiple sim-

ulations and then choose a single action to perform.

Between each simulation, the Evaluate outcome sub-

system must reinitialize the MAS to its initial state

to be ready to perform a new simulation; this is the

main role of the Rollback service. After choosing the

action, the sub-system applies a Commit to validate

this action. The commit re-initializes the MAS to its

last state that corresponds to the environment (before

the simulation) and informs AS of the chosen plan of

action in order to compare the real results obtained

during the last cycle with the simulation ones.

4.1.3 Select Action

This module includes a set of procedures describ-

ing simple actions. Each action represents a robot’s

movement. It is generally used to perform a simple

task, such as moving forward, turning left or right,

pushing an object, and so on. These actions are hard

coded, i.e. the system can not enrich the possible

actions. This module serves as a knowledge base

of robot’s actions; for example the module Evaluate

Outcome wants execute an action such as ”go for-

ward”, then it does not need to know how to do this

action. It need only call a ”go forward” routine, the

details of the implementation of this routine is de-

scribed in this module.

4.1.4 Evaluate Outcome

The objective of this module is to provide a plan of

action. It constructs an oriented and weighted graph

between the current position of the robot and the goal,

the graph is obtained using information from AS. In

VICA implementation, heuristics on the ﬁrst-level of

the graph correspond to effort (distance + effort if

the robot must push an obstacle). Heuristics on dis-

tance are obtained through the distances provided by

LIDAR and the depth camera, heuristics concerning

the effort needed is obtained by simulation. Heuris-

tics on the other levels correspond to the linear dis-

tance between the node and the goal. Whenever the

robot advances the heuristics on the ﬁrst level of the

tree are recalculated, and the distances are readjusted

if necessary. The choice of a plan of action is ob-

tained by the A∗ algorithm

. The system chooses

a branch to explore and performs simulation to up-

date the heuristics. The process is repeated until all

ﬁrst-level branches have been explored. This phase

produces a new graph that considers the effort needed

to move the obstacle. The new graph will be used to

determine the path to be taken by A∗. Of course it is

possible to use other path search algorithms, for now

we have only experimented A∗.

In our case, we use two heuristics, the estimated cost

to overcome the closest obstacle that hinders the passage,

which is represented by the ﬁrst level of the graph, and the

distance that separates the robot from the goal, in the rest of

the levels. When the robot advances towards to the goal, the

graph is reduced, so the second level becomes the ﬁrst and

so on.

ICAART 2021 - 13th International Conference on Agents and Artiﬁcial Intelligence

302

5 IMPLEMENTATION AND

EXPERIMENTS

The goal of VICA is to enable the robot to make a

plan and adapt it, if necessary, during its route, to

reach the goal. The plan may involve to a) move ob-

stacles in order to pass, and b) ask another agent (for

example, another person or robot blocking its path) to

let it pass. A ﬁrst version of VICA was implemented

for a virtual robot in order to evaluate the behaviour

produced. It was tested in a scenario that presented a

cluttered environment including different obstacles of

varying size.

Below, we present an example of simple scenario

describing the robot confronted with an environmen-

tal conﬁguration in which the goal is difﬁcult to reach

without moving the objects in its path (cf Fig 2). (1)

The module responsible for perception observes the

scene and produces a representation in a multi-agent

system (MAS). (2) The module tasked with actions

produces different plans of action in the form of a

graph. (3) All action plans are simulated: push the

cube, push the cylinder or pass between the two. The

robot is equipped with a force sensor. During the sim-

ulation, the force required to move each obstacle is

recorded. After simulation, the heuristics (the force

necessary to move an obstacle) in the graph are up-

dated, and a path is calculated with help from algo-

rithm A*. In the simulation Fig.2, the action plan be-

tween the two objects is identiﬁed as the path with the

least resistance. (4) Finally, the chosen plan of action

is applied to the environment.

The graph (cf. Fig.3) shows the possibilities of ac-

tions in the previous environment conﬁguration pre-

sented in Fig.2. Each node can represent an obstacle

that the robot considers able to move, or a path be-

tween two obstacles to reach a position closer to the

goal (G). The robot uses heuristics (H) that obtain via

simulations that will be used to run the A* algorithm.

Figure 2: Example of random conﬁguration.

The Table 1 compares our VICA model with two

other well-known RRT and D* Lite planning meth-

ods. VICA is able to move obstacles, the number of

movements is indicated in the column (Movements).

The execution time of the movement is indicated in

seconds in the (Time) column. The columns (dis-

Figure 3: Generated graph according to the observed envi-

ronment.

tances) indicate the length traveled in centimeters.

The distance between the starting position and the

goal is ﬁxed (350 cm) for all the experiments. Our

method is able to do better than D *, because it is

able to move obstacles and ﬁnd a better path instead

of avoiding obstacles. The experiments shown here

only involve two types of obstacles (ﬁxed and mov-

able), interactive obstacles (robots, humans, etc.) are

not tested.

VICA models human-like navigation behaviors

when facing obstacles. Often the human brain when

seeking solutions, chooses solutions that require the

least effort. For example, if we wish to pass to the

other side either going around the table, or moving the

chair that hinders the passage, the second solution is

often chosen. The solution proposed by VICA gives

results close to human behavior. The implemented

learning process forces the robot to interact with oth-

ers to learn and increase its ability to respond to sim-

ilar situations. After failing to move unmovable ob-

jects, the robot considers other actions. This process

is similar to natural human behavior.

6 CONCLUSION AND

PERSPECTIVES

In this article, we described VICA, a VIcarious Cog-

nitive Architecture applied to a mobile robot operat-

ing in a crowded environment (NAMO). We produced

a model with MAS representation of the environment,

where the robot is able to inform and understands the

evolution of it, while acting and changing its behavior

appropriately. The assumption of how the environ-

ment could offer to the robot the knowledge gained

through interactions, was conﬁrmed by the results that

show the model’s efﬁciency. VICA offers an intelli-

gent guide to the robot to perform tasks among mo-

bile agents, by avoiding a maximum number of ob-

stacles while reducing the computation time. This

architecture is further validated by interactions with

more complex objects (eg men, other robots, etc.) and

complex scene conﬁgurations to verify that the robot

is able to evolve in a truly complex and natural envi-

VICA: A Vicarious Cognitive Architecture Environment Model for Navigation Among Movable Obstacles

303

Table 1: The results of simulation.

VICA RRT D* Lite

Nbr of obstacles Nbr of movements Travel time (sec) Distance traveled (cm) Distance Distance

5 0 112 370 418 365

10 2 118 388 436 387

20 4 152 385 444 380

30 7 142 358 465 485

40 9 145 390 487 498

50 10 190 378 395 395

60 12 180 397 409 404

70 13 145 395 415 411

80 14 170 412 - -

90 15 186 460 - -

100 15 178 489 - -

ronment. Further work on knowledge representation

is necessary to be compatible with different types of

entities evolving in the environment. The multimodal

perception module will be completed for the extrac-

tion of all the possibilities of actions on an object

(Prospects). Improving the learning module with a

gradually dynamic knowledge would ensure the best

conﬁguration for the goal in any environment.

REFERENCES

Anderson, J. R. (2019). Cognitive architectures including

act-r. Cognitive Studies: Bulletin of the Japanese Cog-

nitive Science Society, 26(3):295–296.

Baars, B. J. (2005). Global workspace theory of conscious-

ness: toward a cognitive neuroscience of human expe-

rience. Progress in brain research, 150:45–53.

Berthoz, A. (2000). The brain’s sense of movement, vol-

ume 10. Harvard University Press.

Berthoz, A. (2008). Neurobiology of” Umwelt”: How Liv-

ing Beings Perceive the World. Springer Science &

Business Media.

Berthoz, A. (2012). Simplexity: Simplifying principles for

a complex world (g. weiss, trans.) cambridge.

Berthoz, A. (2017). The vicarious brain, creator of worlds.

Harvard University Press.

Berthoz, A. and Debru, C. (2015). Anticipation et

ediction: du geste au voyage mental. Odile Jacob.

BERTHOZ Alain, T. M.-H. (2015). Towards creative vi-

cariance. Presses Universitaires de Vincennes, Revue

Hybrid, 2:1–6.

Chen, P. C. and Hwang, Y. K. (1990). Practical path plan-

ning among movable obstacles. Technical report, San-

dia National Labs., Albuquerque, NM (USA).

Choi, D. and Langley, P. (2018). Evolution of the icarus

cognitive architecture. Cognitive Systems Research,

48:25–38.

Djerroud, H. and Cherif, A. A. (2018). Visualization tool

for jade platform (jex). In Proceedings of the Future

Technologies Conference, pages 481–489. Springer.

Djerroud, H. and Cherif, A. A. (2019). Environment engine

for situated mas. In ICAART (1), pages 129–137.

Friedlander, D. and Franklin, S. (2008). Lida and a theory

of mind. Frontiers in Artiﬁcial Intelligence and Appli-

cations, 171:137.

Kotseruba, I., Gonzalez, O. J. A., and Tsotsos, J. K. (2016).

A review of 40 years of cognitive architecture re-

search: Focus on perception, attention, learning and

applications. arXiv preprint arXiv:1610.08602, pages

1–74.

Laird, J. E. (2012). The Soar cognitive architecture. MIT

press.

Lemaignan, S., Ros, R., Alami, R., and Beetz, M. (2011).

What are you talking about? grounding dialogue in a

perspective-aware robotic architecture. In 2011 RO-

MAN, pages 107–112. IEEE.

Levihn, M., Scholz, J., and Stilman, M. (2013). Planning

with movable obstacles in continuous environments

with uncertain dynamics. In 2013 IEEE International

Conference on Robotics and Automation, pages 3832–

3838. IEEE.

Martin, L., Jaime, K., Ramos, F., and Robles, F. (2020).

Declarative working memory: A bio-inspired cog-

nitive architecture proposal. Cognitive Systems Re-

search.

McIntyre, J., Zago, M., Berthoz, A., and Lacquaniti, F.

(2001). Does the brain model newton’s laws? Na-

ture neuroscience, 4(7):693–694.

Mirabel, J. and Lamiraux, F. (2016). Constraint graphs:

Unifying task and motion planning for navigation and

manipulation among movable obstacles.

Moghaddam, S. K. and Masehian, E. (2016). Planning robot

navigation among movable obstacles (namo) through

a recursive approach. Journal of Intelligent & Robotic

Systems, 83(3-4):603–634.

Mueggler, E., Faessler, M., Fontana, F., and Scaramuzza,

D. (2014). Aerial-guided navigation of a ground

robot among movable obstacles. In 2014 IEEE Inter-

national Symposium on Safety, Security, and Rescue

Robotics (2014), pages 1–8. IEEE.

Okada, K., Haneda, A., Nakai, H., Inaba, M., and Inoue,

H. (2004). Environment manipulation planner for hu-

manoid robots using task graph that generates action

ICAART 2021 - 13th International Conference on Agents and Artiﬁcial Intelligence

304

sequence. In 2004 IEEE/RSJ International Confer-

ence on Intelligent Robots and Systems (IROS)(IEEE

Cat. No. 04CH37566), volume 2, pages 1174–1179.

IEEE.

Rao, A. S. and Georgeff, M. P. (1991). Modeling rational

agents within a bdi-architecture. KR, 91:473–484.

Remmelzwaal, L. A., Mishra, A. K., and Ellis, G. F.

(2020). Brain-inspired distributed cognitive architec-

ture. arXiv preprint arXiv:2005.08603.

Renault, B., Saraydaryan, J., and Simonin, O. (2019). To-

wards s-namo: socially-aware navigation among mov-

able obstacles. In Robot World Cup, pages 241–254.

Springer.

Sabah, G. and Briffault, X. (1993). Caramel: A step towards

reﬂection in natural language understanding systems.

In Proceedings of 1993 IEEE Conference on Tools

with Al (TAI-93), pages 258–265. IEEE.

Stilman, M. and Kuffner, J. (2008). Planning among mov-

able obstacles with artiﬁcial constraints. The Interna-

tional Journal of Robotics Research, 27(11-12):1295–

1307.

Sun, R. (2006). From cognitive modeling to social simu-

lation. Cognition and multi-agent interaction: From

cognitive modeling to social simulation, page 79.

Sun, Y., Ren, S., and Lin, Y. (2014). Object–object inter-

action affordance learning. Robotics and Autonomous

Systems, 62(4):487–496.

Trullier, O., Wiener, S. I., Berthoz, A., and Meyer, J.-A.

(1997). Biologically based artiﬁcial navigation sys-

tems: Review and prospects. Progress in neurobiol-

ogy, 51(5):483–544.

Vernon, D., Von Hofsten, C., and Fadiga, L. (2011).

A roadmap for cognitive development in humanoid

robots, volume 11. Springer Science & Business Me-

dia.

Wexler, M., Kosslyn, S. M., and Berthoz, A. (1998). Motor

processes in mental rotation. Cognition, 68(1):77–94.

Wooldridge, M. (2009). An introduction to multiagent sys-

tems. John Wiley & Sons.

Wu, H.-n., Levihn, M., and Stilman, M. (2010). Navigation

among movable obstacles in unknown environments.

In 2010 IEEE/RSJ International Conference on Intel-

ligent Robots and Systems, pages 1433–1438. IEEE.

Ye, P., Wang, T., and Wang, F.-Y. (2018). A survey of cog-

nitive architectures in the past 20 years. IEEE trans-

actions on cybernetics, 48(12):3280–3290.

VICA: A Vicarious Cognitive Architecture Environment Model for Navigation Among Movable Obstacles

305