The Missing Tip: Lack of Micro-Movements Impairs Navigation Realism

in Artiﬁcial Social Agents

Jacob Sharp

and Ulysses Bernardet

College of Engineering and Physical Sciences, Aston University, Aston St, Birmingham, U.K.

Keywords:

Artiﬁcial Social Agents, Navigation, Micro-Movements, Virtual Environments, Path Planning, Realism,

Immersion, Standardisation.

Abstract:

Navigation is critical to an intelligent social agent’s ability to interact with the world and any other agent,

virtual or otherwise. In order to create a truly realistic artiﬁcial social agent, unconscious human micro-

movements need to be simulated. We see this as an important goal for the research area. Examples of these

micro-movements include orienting while walking and back-stepping, straﬁng with attention focused else-

where, and micro-orientations during locomotion. We postulate that there is a gap in research around these

micro-movements within the ﬁeld of navigation that we hope to contribute to ﬁlling. Most research in this ﬁeld

is focused on the understandably important pathﬁnding aspect of navigation; moving between two spatial lo-

cations. There is little to no research being done on micro-movements and making a truly realistic navigation

system for artiﬁcial social agents. Moreover, there exists no canonical way of describing these movements

and ”micro-movements” that are so characteristic for human spatial behaviour. Here we propose a set of stan-

dardised descriptors of movement conﬁgurations, that will be able to be used as building blocks for spatial

behaviour experimentation, and as the basis for behaviour generation models. We see this as an important tool

in the creation of navigation systems that are able to more readily include these kinds of behaviours, with hope

that the aforementioned conﬁgurations will improve development of realistic movement systems.

1 INTRODUCTION: HIERARCHY

OF NAVIGATIONAL REALISM

The navigational behaviour of an artiﬁcial social

agent plays a key role in the perception of realism hu-

mans feel when interacting with it. However, much

current research into this area is focused primarily

on allowing intelligent artiﬁcial social agents to navi-

gate to their intended destination realistically, or to in-

crease the visual ﬁdelity of the artiﬁcial social agents

themselves. This has created a gap in the litera-

ture with little to no research done into the micro-

movements and orientation that we believe give an

artiﬁcial social agent a much higher sense of real-

ism and therefore foster a higher sense of immer-

sion in the user; the subjective feeling of being in an-

other world (Bartle, 2004). This is important in vir-

tual experiences as it has been shown that a higher

sense of immersion is key in creating effective vir-

tual worlds. With virtual experiences being used

https://orcid.org/0000-0002-5577-3962

https://orcid.org/0000-0003-4659-3035

more often for important experiences such as train-

ing (Lele, 2013), (Merchant et al., 2014), the onus

is there to create immersive worlds in which the ex-

perience that humans gain is real and worthwhile.

(Latoschik et al., 2017) conﬁrmed this with their work

on avatar realism and it’s effect on various measures,

ﬁnding that humaneness and attractiveness were both

increased with human-like avatars over wooden man-

nequins, though eeriness was also increased. These

ﬁndings were supported by (P

utten et al., 2009) look-

ing into whether social presence can be elicited for

virtual avatars, with results showing that they can,

with the more behavioural realism the artiﬁcial social

agent shows increasing this measure. Other than be-

havioural realism, movement is another key area of

the realism of an artiﬁcial social agent, though it can

be seen as one of the most important, (Pedica and

Vilhj

almsson, 2008) found from their work on vir-

tual avatars for online chat rooms that social percep-

tion and reactive manoeuvring in the form of group-

ing around o-spaces (Burgoon and Kendon, 1992)

seemed to give the avatars a heightened level of re-

alism showing that the orientation and positioning

380

Sharp, J. and Bernardet, U.

The Missing Tip: Lack of Micro-Movements Impairs Navigation Realism in Artiﬁcial Social Agents.

DOI: 10.5220/0011798300003393

In Proceedings of the 15th International Conference on Agents and Artiﬁcial Intelligence (ICAART 2023) - Volume 1, pages 380-386

ISBN: 978-989-758-623-1; ISSN: 2184-433X

 2023 by SCITEPRESS – Science and Technology Publications, Lda. Under CC license (CC BY-NC-ND 4.0)

of people or agents can be just as important. This

need for realistic movement can be seen again in the

propagation of motion capture technology in many

forms of media. This high cost, high time invest-

ment form of animation, uses recorded humans to cre-

ate realistic movements for virtual characters. The

capture of the micro-movements and unconscious ac-

tions is what make motion capture so powerful, if

there were a way to simulate these micro-movements

without the heavy cost of motion capture, the real-

ism and believability of artiﬁcial social agents could

increase without a heavy investment. This in turn

would increase the overall immersion of the expe-

rience (Pedica and Vilhj

almsson, 2008). However

the main focus of pathﬁnding, determining the path

the agent takes from (a) to (b) is understandable, as

for these micro-movements and navigations to occur,

there must be a movement scenario for them to occur

within. Thus, the construction of a believable human

movement model can be viewed as a sort of pyramid

(Figure 1). Inspired in part by A. Maslow’s hierar-

chy of human needs (Maslow, 1943), with pathﬁnd-

ing laying the foundational systems for higher level

behavioural systems to utilise. We see the micro-

movement behaviours at the ”tip” of the pyramid that

are described in this paper as being supremely impor-

tant in the development of realistic human artiﬁcial

social agent, and that these behaviours, small move-

ments, navigation and orientation all have a very im-

portant part to play in creating a realistic, immersive

artiﬁcial social agent that evidences some facsimile

of life (Pedica and Vilhj

almsson, 2008). For humans,

these things happen subconsciously without thought

(Burgoon and Kendon, 1992) but for artiﬁcial social

agents, these need to be ﬁrstly measured from human

participants, canonised and studied and then designed

in such a way for an AI to perform these small actions.

The current article addresses the scientiﬁc commu-

nity, speciﬁcally around artiﬁcial social agents. It’s

goal is to increase the awareness of the shortcomings

in the area of locomotion for artiﬁcial social agents,

and realism in virtual worlds. The research areas

are introduced ﬁrst, with Pathﬁnding, Social Robotics

and Micro-Movements explored in more detail. Ex-

amples of current locomotion data will be presented

to illustrate the issue with current locomotion data

collection, before moving onto the proposal of this

paper; the presentation of the formalisation of these

micro-movements we unconsciously make during lo-

comotion.

Figure 1: The requirements of a realistic human artiﬁcial

social agent as a hierarchical pyramid; Pathﬁnding describ-

ing the routing method that generates the path for an arti-

ﬁcial social agent to take. Locomotion is the actual move-

ment the agent employs to displace itself to the intended

direction. Micro-Movements are the small movements and

orientations humans use during navigation such as rotating

while stepping, that when lacking cause a lowering of real-

ism of the agent.

1.1 Pathﬁnding - Getting from (a) to (b)

Pathﬁnding as a subset of navigation has a large body

of research due to it’s necessity regarding the control

models of real-world robots, however this robotic fo-

cus has led to most systems having very low mea-

sures of realism when applied to an artiﬁcial social

agent. When investigating multi agent systems (Wang

et al., 2013) focused on creating a dynamic system

for controlling multiple agents in an immersive en-

vironment, with special attention on path planning

for avoidance. The purpose of this research was to

increase the effectiveness of training simulations for

evacuation of multiple agents and other real-time ap-

plications of crowd simulation, however misses the

importance of increasing the realism of the individ-

ual movement of each agent. (Zhukov and Iones,

2000)’s work on navigational control for intelligent

agents instead focused on the creation of navigational

maps, with the goal of decreasing computational load

and increasing the complexity of navigation for artiﬁ-

cial social agents, but still only designed these nav-

igational maps to translate an agent from point (a)

to point (b) with no simulation of higher movement

functions. With (Raees and Ullah, 2021) also focus-

ing on pathﬁnding without the complexity of micro-

movements we discuss in this paper. (Olcay et al.,

2020) Took the extremely interesting angle of de-

signing a simultaneous, collision-free motion plan-

ning system for fully autonomous robots, allowing

groups of fully autonomous robots to motion-plan,

even within an environment with moving obstacles

The Missing Tip: Lack of Micro-Movements Impairs Navigation Realism in Artiﬁcial Social Agents

381

or poor sensor range. This kind of SLAM (simulta-

neous localisation and mapping) control system will

be hugely important as a facet of an autonomous true

virtual human. However for the goal of creating com-

puter controlled artiﬁcial social agents which move in

a realistic way, autonomy is not a requirement. Re-

inforcing the research into navigation within an envi-

ronment with movable obstacles, (Djerroud and Ali-

erif, 2021) created ”VICA” a vicarious cognitive

architecture for autonomous robots, though this re-

search differs as it follows the ”theory of mind” in

saying that a form of ”vicariance” is important for

a robot’s strategy to interact with the outside word.

This architecture employs a multi agent system to al-

low the robot a representation of how it’s interactions

would cause the outside world to react. (Sutera et al.,

2021) have pushed the ﬁeld of marrying navigation

with learning even further, by using ultra-wide band

technology for precise tracking combined with a low-

cost point-to-point local planner learnt with deep re-

inforcement learning (the notion of intelligent agents

taking actions to maximise a cumulative reward, see

(Akalin and Loutﬁ, 2021)), they are able to path-ﬁnd

robustly in noisy and complex environments. This

is something important for robots in real-world envi-

ronments, but unnecessary for artiﬁcial social agents,

who by virtue of their medium already have access to

all data on their environment. These approaches once

again however, all miss a vital component of realism

in these artiﬁcial social agents, the micro behaviours

that we propose need formalisation.

1.2 Approaches in the Social Robotics

Domain

Robotics has a wealth of valuable data on naviga-

tion due to the ﬁeld of Human-robot interaction (HRI)

being one of the largest in artiﬁcial social agents,

with increasing amounts of research being done in the

area due to the uptake of complex and use-speciﬁc

robots that exist in the world today (Goodrich and

Schultz, 2007). One may think this domain has re-

search into micro-movements due to it’s size, how-

ever even though the ﬁeld of HRI is so large, the

most related research in this ﬁeld is focused primar-

ily on interactive social robots. For example (Ghaz-

ali et al., 2019) Looked at the effect of social cues

in robots on user’s psychological reactance, liking

among other psychological measures, however they

do not investigate navigational realism as a social

cue, instead focusing only head mimicry and social

praise timing. (Liu et al., 2018) investigated human-

robot behaviour in a shopkeeper scenario and in-

cluded locomotion in the multi-model behaviour of

the robot. Finding that cross-validation on the train-

ing data showed higher social appropriateness of the

robot’s behaviours. Though once again this research

was conducted on a wheeled form of locomotion, and

realistic human movement was not the intention of

the research. Apart from investigations into social

robots such as these, the bulk of research is directed

around navigation systems that create efﬁcient (a) to

(b) routes for robotic agents (Olcay et al., 2020), (Li

et al., 2019). Despite the ﬁeld of HRI being so large,

it is still missing research into the micro-movements

that we describe in this paper. This is primarily due

to robotics in general not being advanced enough in

their mimicry of human movement in a reliable way

to focus on these higher-level behaviours, causing the

area of advanced realism of movement to be some-

thing that will need to be researched in the future.

2 HOW DO HUMANS NAVIGATE?

2.1 Micro-Movements - The Small

Movements We Make

Research into the small movements we unconsciously

make is an even smaller subset of the navigation ﬁeld

and is ongoing, and these behaviours give a true sense

of realism when simulated or replicated well, though

most of this research is focused on face or arm move-

ments, rather than the implicit orientations and move-

ments we make during locomotion. These locomotion

movements are what we hope to formalise, as work

on facial realism and animation realism is quite ma-

ture in comparison due to the important role the face

has for humanoid perception, as well as due to mo-

tion capture as a technology. For example (Davison

et al., 2018) in their work in the ﬁeld of micro fa-

cial movements have created a formalised dataset of

micro-facial movements, poised to become the new

standard. No such dataset exists for this kind of data

relevant to human navigational movements however,

and this sort of gap in the ﬁeld explains why even in

highly funded, yet unreleased video games such as

”Star Citizen” (see (Ahrens et al., 2019)) non player

character performance is still substandard, and often

consists of movement to a position before rotating and

continuing a path. (Onishi et al., 2003) investigated

creating a new laboratory application to record hu-

man robot movements and test new humanoid robots,

that describe in their future work section the need to

capture human locomotion data accurately to make

robots that realistically move like humans. This is a

good example of the need for these realistic move-

ments being recognised, with work being done to im-

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

382

prove the data capture systems that are used to cre-

ate spatial control systems which once again conﬁrms

the necessity for further research into agents’ naviga-

tion if the aim is to increase the ”sense of realism”.

(Kuffner et al., 2003) have done incredibly impor-

tant work in the ﬁeld, with their goal of creating a

digital human, including sophisticated digital models

of human physiology and biomechanics. However at

least at this stage, they have been more focused on

arm motion and upper body movements at a set lo-

cation, rather than the movements we describe here.

(Kagami et al., 2003) looks at using a motion cap-

ture system, force plates and distributed force sensors

to record data from a human participant, as well as a

humanoid ”H7” robot, however they found difﬁculty

comparing the human data with the H7 due to issues

in differences between the mechanical nature of a hu-

man and a robot, (including link parameters, walk-

ing speed, step length, step cycle and mechanisms).

These issues however do not exist in the virtual world

unless explicitly designed, and the large amount of

heavy data Kagami recorded emphasises the need for

a formalisation of these naturalistic, realism focused,

micro-movements. Even (Gratch et al., 2002) in their

excellent paper on virtual human realism fail to men-

tion navigation in any form, they note that the broad

range of requirements virtual humans have poses a

serious problem for research and the technology in

general. But do not touch on the speciﬁcs of micro-

movement that we describe here.

2.2 Human Movement Data

Human navigational behaviour is complex and the

data required to fully understand it can have a lot of

nuances, due to the innate complexity of the move-

ment it is representing. When looking at tracked data

from a human in an environment, such as the illustra-

tion in Figure 2 it is clear a simple location tracking of

the participant gives a good representation of where

the human moved, but not in what order, or where

they were facing during the movement. If one was

developing a human artiﬁcial social agent control sys-

tem with an emphasis on realism, data like this may

help in creating the pathﬁnding control, but this data

is not able to be easily utilised in the creation of a con-

trol model that includes micro-movements. When the

shoulders of the participant are tracked as in Figure

3, this gives a much better sense of where the human

was facing during locomotion and can even convey

a sense of turning while moving and straﬁng. This

begins to highlight the complexity of human locomo-

tion, as humans do not just face forward and move

along a straight axis, and a better way to describe the

different parts of this data would go a long way in in-

creasing the ease of implementing micro behaviours

in artiﬁcial social agents.

Figure 2: Centre of mass of a person moving through space

between 4 different spatial locations.

When looking at the illustrative human accelera-

tion data in Figure 4, one can once again see human

locomotion is not as simple as moving to a location

and stopping to turn and continuing along the calcu-

lated path, acceleration is more than just a linear up

and down. The data instead includes a large array

of different movements and angular velocities, further

highlighting how complex this data can become, and

again showing that humans indeed do employ micro-

movements that if simulated can improve the realism

of an agent, are so often missed out during research

into navigation and are the exact movements we pro-

pose need to be formalised. With annotation of the

human navigational data (Figure 3, one can see that

each part of the movement data can be ascribed to a

speciﬁc micro-movement, though even the rotate in

place annotated here may be closer to rotate in walk.

These micro-movements are described in more detail

within the next section.

This issue of translating human movement into the

virtual space has been wrestled with for a long time,

and though we do not yet have a solution for this prob-

lem, we do propose a framework for describing and

formalising this behaviour and improving ease of col-

lecting useful data.

3 FORMALISING

MICRO-MOVEMENT

BEHAVIOUR

To have realistic agent behaviour, we need to de-

velop spatial models that understand these micro be-

The Missing Tip: Lack of Micro-Movements Impairs Navigation Realism in Artiﬁcial Social Agents

383

Figure 3: Orientation during locomotion between 4 dif-

ferent spatial locations, with speciﬁc micro-movement be-

haviours from the proposed formalisation annotated.

Figure 4: These Plots of Orientation, distance and velocity

show that the orientation of a human is constantly changing

and complex in it’s nature, there is a lot of changes in accel-

eration, velocity and orientation rather than a simple (a) to

(b) with constant velocity.

haviours, and to that end this paper proposes the for-

malisation of a testing framework, by associating the

following descriptors to these micro behaviours with

diagrams representing the 6 behaviours in Figure 5.

If we assume that there are a ﬁnite set of movement

strategies that humans will use, one can divide human

locomotion into 6 key behaviours. We see these be-

haviours as consisting of; (A) ”Linear walk,” where

an agent walks in a direct line congruent with their

orientation direction. (B) ”Backwards walk,” where

an agent back-steps without turning their orientation.

while facing forward. (D) ”Curved walk,” where an

agent is orienting gradually as they walk. (E) ”Rotate

in Place,” where an agent may turn on the spot without

movement, and ﬁnally (F) ”Rotate in Walk,” in which

an agent is rotating their orientation as they step in a

speciﬁc direction, for example, turning as you back-

step into a forward walk. One can view any naviga-

tion from point (a) to point (b) as a combination of any

number of these actions, resulting in the ability for in-

vestigations into this ﬁeld to more readily explain and

describe the movement behaviours they record. At

the very least, it is important for the ﬁeld of agent

navigation to recognise these important navigational

micro-movements, as they are paramount in creating

the next step in virtually real humanoid agents that

foster a sense of immersion in users.

4 WHERE TO NEXT?

From the research discussed above, it is clear that

there exists a gap within the ﬁeld of navigational

research centred around the simulation of micro-

movements that humans unconsciously employ dur-

ing path-planning and execution of movement. Be-

haviours such as these have been found to foster

a higher sense of immersion within virtual worlds,

something that is key to creating impactful, efﬁcient

virtual simulations. This is especially important in

the ﬁeld of virtual reality training, as the closeness to

the real-world is what makes it such a powerful tool.

With the formalisation of these micro-movements that

occur during locomotion, researchers will be able

to look at complex human locomotion data and as-

cribe these formalised terms to the different micro-

movement techniques that are employed. This will

result in researchers being more well equipped to

tackle the issue of realistic human movement in a vir-

tual space, creating this much needed and oft missing

sense of realism to the artiﬁcial social agents that in-

habit these virtual spaces. Formalisation is just the

ﬁrst step however, and this future work must look at

collecting data on these micro-movements during lo-

comotion from real-life human participants, and cre-

ate an autonomous system that is able to replicate in

a realistic way and simulate these micro-movement

behaviours for an artiﬁcial social agent. Whether by

reinforcement learning systems, ﬁnite control mod-

els or some other system. This is speciﬁcally for AI

controlled agents in virtual spaces, as user-controlled

agents do not run into this issue as the user is in full

control of the navigation and thus any strange move-

ment is overlooked. The list that has been outlined

here can be seen as complete, though is open to being

improved upon, and is a starting point for the stan-

dardisation of these concepts for ease of understand-

ing and formalisation.

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

384

Figure 5: Proposed list of atomic micro-movements, complete in that we posit there is no further way to break these motions

down in a meaningful way. The start location of the agent is the lighter blue circle, the darker blue circle denotes the end

location of movement. The arrows on the circle show body orientation, the yellow arrows denote the displacement path

irrespective of body orientation. (A) Linear walk in a direct line congruent with body orientation. (B) Linear back-step in a

direct line. (C) Strafe (side-step) to the side while facing forward. (D) Curved walk with gradual body orientation along the

curve. (E) Rotating in place with no translation of the center of mass of the body, the original and ﬁnal orientation differ by

180

◦

. (F) Rotation during displacement along a path, the original and ﬁnal orientation also differ by 180

◦

REFERENCES

Ahrens, J. P., Istipliler, B., Isaak, A., and Steininger, D. M.

(2019). The star citizen phenomenon & the “ulti-

mate dream management” technique in crowdfunding.

In 40th International Conference on Information Sys-

tems, ICIS 2019.

Akalin, N. and Loutﬁ, A. (2021). Reinforcement learning

approaches in social robotics.

Bartle, R. A. (2004). Designing Virtual Worlds.

Burgoon, J. K. and Kendon, A. (1992). Conducting Inter-

action: Patterns of Behavior in Focused Encounters.

Contemporary Sociology.

Davison, A. K., Lansley, C., Costen, N., Tan, K., and Yap,

M. H. (2018). SAMM: A Spontaneous Micro-Facial

Movement Dataset. IEEE Transactions on Affective

Computing.

Djerroud, H. and Ali-Ch

erif, A. (2021). VICA: A vicari-

ous cognitive architecture environment model for nav-

igation among movable obstacles. ICAART 2021 -

Proceedings of the 13th International Conference on

Agents and Artiﬁcial Intelligence, 2(Icaart):298–305.

Ghazali, A. S., Ham, J., Barakova, E., and Markopoulos,

P. (2019). Assessing the effect of persuasive robots

interactive social cues on users’ psychological reac-

tance, liking, trusting beliefs and compliance. Ad-

vanced Robotics.

Goodrich, M. A. and Schultz, A. C. (2007). Human-robot

interaction: A survey. Foundations and Trends in

Human-Computer Interaction, 1(3):203–275.

Gratch, J., Rickel, J., Andre, E., Cassell, J., Petajan, E.,

and Badler, N. (2002). Creating Interactive Virtual

Humans: Some Assembly Required. IEEE Intelligent

Systems.

Kagami, S., Mochimaru, M., Ehara, Y., Miyata, N., Nishi-

waki, K., Kanade, T., and Inoue, H. (2003). Measure-

ment and comparison of human and humanoid walk-

ing. In Proceedings of IEEE International Sympo-

sium on Computational Intelligence in Robotics and

Automation, CIRA.

Kuffner, J., Chestnutt, J., Kagami, S., Latombe, J. C., Nishi-

waki, K., Hodgins, J., LaValle, S., Inaba, M., Yamane,

K., and Inoue, H. (2003). Motion planning for digital

humans. Proceedings of IEEE International Sympo-

sium on Computational Intelligence in Robotics and

Automation, CIRA, 2:912–917.

Latoschik, M. E., Roth, D., Gall, D., Achenbach, J., Walte-

mate, T., and Botsch, M. (2017). The effect of avatar

realism in immersive social virtual realities. Proceed-

ings of the ACM Symposium on Virtual Reality Soft-

ware and Technology, VRST, Part F1319.

Lele, A. (2013). Virtual reality and its military utility. Jour-

nal of Ambient Intelligence and Humanized Comput-

ing, 4(1):17–26.

Li, R., Van Almkerk, M., Van Waveren, S., Carter, E.,

and Leite, I. (2019). Comparing Human-Robot

Proxemics between Virtual Reality and the Real

World. ACM/IEEE International Conference on

Human-Robot Interaction, 2019-March:431–439.

Liu, P., Glas, D. F., Kanda, T., and Ishiguro, H. (2018).

The Missing Tip: Lack of Micro-Movements Impairs Navigation Realism in Artiﬁcial Social Agents

385

Learning proactive behavior for interactive social

robots. Autonomous Robots.

Maslow, A. H. (1943). A theory of human motivation. Psy-

chological Review.

Merchant, Z., Goetz, E. T., Cifuentes, L., Keeney-

Kennicutt, W., and Davis, T. J. (2014). Effectiveness

of virtual reality-based instruction on students’ learn-

ing outcomes in K-12 and higher education: A meta-

analysis. Computers and Education.

Olcay, E., Schuhmann, F., and Lohmann, B. (2020). Col-

lective navigation of a multi-robot system in an un-

known environment. Robotics and Autonomous Sys-

tems, 132:103604.

Onishi, M., Odashima, T., Asano, F., and Luo, Z. (2003).

Development of PC-based 3D dynamic human inter-

active robot simulator. In Proceedings of IEEE Inter-

national Symposium on Computational Intelligence in

Robotics and Automation, CIRA.

Pedica, C. and Vilhj

almsson, H. (2008). Social perception

and steering for online avatars. Lecture Notes in Com-

puter Science (including subseries Lecture Notes in

Artiﬁcial Intelligence and Lecture Notes in Bioinfor-

matics), 5208 LNAI:104–116.

utten, A. M. V. D., Kr

amer, N. C., and Gratch, J. (2009).

Who’s there? Can a Virtual Agent Really Elicit Social

Presence? Design, pages 1–7.

Raees, M. and Ullah, S. (2021). RUN: rational ubiqui-

tous navigation, a model for automated navigation and

searching in virtual environments. Virtual Reality,

25(2):511–521.

Sutera, E., Mazzia, V., Salvetti, F., Fantin, G., and Chi-

aberge, M. (2021). Indoor point-to-point navigation

with deep reinforcement learning and ultra-wideband.

In ICAART 2021 - Proceedings of the 13th Inter-

national Conference on Agents and Artiﬁcial Intelli-

gence.

Wang, Y., Dubey, R., Magnenat-Thalmann, N., and Thal-

mann, D. (2013). An immersive multi-agent sys-

tem for interactive applications. Visual Computer,

29(5):323–332.

Zhukov, S. and Iones, A. (2000). Building the navigational

maps for intelligent agents. Computers and Graphics

(Pergamon), 24(1):79–89.

ICAART 2023 - 15th International Conference on Agents and Artiﬁcial Intelligence

386