beaverswimmingΤεχνίτη Νοημοσύνη και Ρομποτική

14 Νοε 2013 (πριν από 4 χρόνια και 8 μήνες)

204 εμφανίσεις






The aim of the project is to establish intelligent autonomous capabilities in service mobile robots

realize a cooperative capabilities in a multi
agent based system. The research aims to investigate the
fundamental issues associated with creating task planning methodologies by interpreting high level
commands to a series of low level capabilities (r
ealized in task primitives) in uncertain environment,
and thus establish intelligent multi
agent based capabilities in a cooperative task. These methodologies
and formulations will be established, investigated and verified not only by simulation but also
using the
mobile robot research facility available in our research laboratory. The specific aims of the project

To investigate techniques in achieving higher level task specification and in interpreting high level
(complex) command, e.g. a comma
nd in human
like language, to a series of lower level task

To utilize, modify, and build upon existing localization, map
building, and navigation techniques to
establish a robust control strategy that allows autonomous mobile robot control in
uncertain and
changing man
made environments.

To investigate fundamental issues associated with multi
agent based systems.

To establish a robust and integrated control strategy in the operation of a complex task in an
intelligent manner through the coope
rative behaviour of multi
agent system.

This proposal is concerned with the immediate and emerging industrial and service sector needs for
autonomous service robots. The industry partner (FloorBotics) associated with this project has already
a need in a variety of national and international industries for such intelligent autonomous
systems. One potential application is industrial cleaning. Pharmaceutical companies and the high
industries operate super
clean factory environments. Currentl
y spillages in these environments require
human intervention. This involves stopping production for up to 48 hours while the air in the
environment, contaminated by the human presence is replaced with clean air. Australian companies,
investing in state
the art factories, require service robots which can both systematically clean these
large spaces and also respond to requests for the quick clean up of spillages. Hospitals and other large
public facilities have already demonstrated a demand for assistive
cleaning and delivery service robots.

To fulfill the above
mentioned requirements, this project focuses on the realization of multi
operation within a finite horizon (indoor application, specifically household or domestic and industrial
(factory) se
ttings) to achieve a higher level complex task. The challenges include human
robot and
robot interaction, task
oriented strategy in obtaining sensor information, sensor fusion, multi
robot task planning and scheduling, multi
robot motion planning, a
nd robust motion control.

Motion control in a mobile robot consists of localisation, map
building and navigation which form the
fundamental capabilities of a mobile robot to explore an environment, known or unknown, and to
achieve a certain objective. Th
ese fundamental capabilities are well
established for indoor
applications. In this project, these capabilities will be utilised and implemented upon existing systems.
Traditionally, localization and navigation of a mobile robot would assume the existence

of pre
knowledge of the environment. More recent research was directed towards solving the problem of
localization and navigation in an unknown area while performing map
building. The challenge in this
effort lies in the fact that localization
and navigation require the knowledge of the environment while
building requires reliable localization capability. This is commonly termed the Simultaneous
Localisation And Map
building (SLAM) problem

(with many variations), in which
much of research effort in this area has been focused in the past decade. Good

reliable stochastic
algorithms have been made available through these research efforts and this project intends to utilize
these in order to facilitate our core research in multi
agent based intelligent autonomous system

agent based system involves the use of multiple homogeneous or heterogeneous agents. In this
project, the mobile robots are utilized a
s the agents. In a homogeneous system, multiple identical
robots are fielded to perform the specified task, while in heterogeneous system, different classes of
robots are fielded

each carrying its own types of sensors, capabilities, and roles. Multi
ent based
systems can also be classified into centralized and decentralized systems. In a centralized system,
processing of data from all other agents is performed by one individual that then distributes the
information accordingly. A decentralized syste
m does not have a central processing agent, but each
agent communicates with each other locally. In an explorative task, challenges lie in the interaction
between the multiple agents and the sharing of information obtained locally by each agent to constru
a globally consistent map of the environment. Multi
agent system is a fast developing area of research
due to the recent rapid development of on
board computing power, embedded systems, and wireless
technologies. Having multiple smaller, simpler and l
ower cost robots to perform a task, compared to
having one large complex robot, offers a more robust and fault tolerant method of achieving a complex
task, as well as speeding up the process through parallel processing of the task and extra degrees of
erity. Some impressive examples of large
scale multi
robot operations have been experimented
with plenty of scope for further fundamental research

The h
robot interaction problem involves the establishment of the strategies to accept high
command, which could be multi
(involving more than one mode of communication), and
to de
compose this task into a series of task primitives
. A task primitive is the most basic
unit of low
level ca
pability of which a series of these could be called in an intelligent manner to
construct a higher level behaviour. Task decomposition in multi
agent system involves the
identification of task primitives as required to form a task, scheduling, allocation,

handling of the task
onto multi
agents, and ensuring the flow and the global completion of a specified high level task
through cooperative effort of multiple agents. This has been attempted through various methods such
as Discrete Event System and Artifi
cial Intelligence
. Challenges also lie in the modeling of
uncertainty involved in the operation of multiple agents, sensory measurement, and dynamic changes

within the environment of interest.


The intelligent autonomous multi
agent based system considered in this project combines the
advantages of cooperative robotics, advanced uncertainty handling methodologies, and adva
nced task
decomposition methodologies. The purpose is to establish advanced planning and control
methodologies for multi
agent systems together with the enabling technology to accomplish
autonomous system capable of cooperatively achieving complex task wi
thin human environment. The
results will find application in industrial environment, from heavy industries to semiconductor clean
room, hospitals, warehouses, even households, and any indoor environment, especially in tasks
inaccessible by or deemed hazar
dous to humans. Tasks performed could include cleaning, delivery,
security application, search
rescue, patrol of large areas, mapping of unknown areas, and many

This research project is uniquely comprehensive in scope and fundamental in charac
ter and is
strategically directed to key issues common to all possible applications requiring multi
agent based
autonomous systems. The expected outcomes of this fundamental research project include:

establishment of an efficient methodology to a process
level command into a series of lower
level commands;

establishment of effective and robust multi
robot task planning and scheduling and multi robot
motion planning;

establishment of an intelligent task
oriented strategy in obtaining sensor informati
on, performing
sensor fusion and uncertainty handling using advanced stochastic methods;

dissemination of research results in refereed international journal and conferences.

This work is significant because it takes a leap ahead in the direction which ex
isting belief state
methods are slowly converging. The research area is clearly significant. The issues being addressed are
fundamental to the deployment of mobile robots for which there is already demand in a variety of
industries. The availability of th
e proposed robot control system and its underlying theory will allow
efficiencies in Australian industry and in public service sectors resulting in either improved quality
services or reduced costs for delivering existing services.

Significance of Floorbo

FLOORBOTICS is a robotic software / hardware development company. It is built around the
researched international market demand for an alternative to the current selection of labour intensive
vacuuming/mop products. FLOORBOTICS concentrates on inn
ovation with strong, visionary
technical expertise. Corporate policy is focused on creating the latest in ‘cutting edge’ set
robotic cleaners at an affordable price. It is dedicated to pioneering design, engineering and electronic

With offices in Australia, the USA, Japan and Europe, FLOORBOTICS is an international enterprise
with many years of comprehensive global experience. Today, it is a world leader in both software and
hardware robotics. Headquartered in Melbourne, Austr
alia and with strategic international links,
FLOORBOTICS aims to realize the existence of a cleaning robotic system in every home, business and


The main focus of this research project is the multi
agent based system
with cooperative behaviour to
accomplish more complex tasks. Basic mobile robot capabilities such as localization, map
and simple navigation are assumed. The first six months of the project will be spent on setting up the
infrastructure for the
project in terms of consolidating current research effort into robust mobile robot
systems capable of localization, map
building, and navigation.

agent based system research is aimed at the development of the governing behaviour of multiple

and its interaction to complete a global task. In this project, the agents are represented by the
mobile robots. The approach would be to represent the mobile robots as reactive/ cognitive agent
rather than a purely reactive one, as purely reactive agen
ts generate intelligence through self
organization of “non
intelligent” agents and lack the flexibility of utilizing a priori knowledge when
they have to perform a different task
. A comparison between centralized and decentralized
configurations would be performed. A centralized configuration has the advantage of having a global
overview of the events, such as knowing the positions of all the agents, and henc
e capable of more
efficient planning for the system.

Knowledge acquiring is performed through various advanced methodologies such as artificial
intelligence and belief
state approaches. These methods take into account the uncertainty and hidden
of the real
world robot interaction with the environment as well as gather representation of the
environment and the high
level task description as communicated by human. It is proposed that a
layer subsumption
like architecture be investigated for
this problem. Level zero will be trained
using reinforcement learning techniques

to navigate generic environments while avoiding
collisions with stationary and moving obstacles and to have basic generic exploration strate
Higher layer is added when the agent is deployed, which extends and customizes the robot’s
representation to a specific deployment environment supplementing, but not replacing, the generic
knowledge. Reinforcement learning also allows a natural and
efficient combination of knowledge from
the two integrated models.
The proposed multi
agent systems can be deployed in an initially unknown
internal environment and use an intelligent mapping strategy which allows them to cooperatively map
that environment

efficiently while moving safely avoiding both moving and stationary obstacles. Map
building and updating is continuous over the lifetime of the agents which will autonomously detect and
reflect changes in the environment.
A strategy would be formulated
to allow the integration of
knowledge obtain by various agents to generate a robust and reliable representation of the larger areas.
Once sufficiently developed, the integrated internal representation of the agents can be depicted in a
form that allows an
notation by a human user to identify locations and regions of interest and to provide
an appropriate labeling. This could be further utilised as one of the multi
modal forms of interaction
between robot and human.

Task specification methods, includi
ng multi
modal means of human
robot interaction will be explored
and an efficient strategy will be established
. Multi
modal human
robot interaction is a new
emerging study that attempts to
make human
robot interaction more intuitive. Several modes of
communications are utilized, such as speech, hand gesture, facial recoginition, mostly methods that are
familiar to human. This methodology of communication is not only more intuitive to human
, but is
also more robust due to the redundancy in the communication channels.

A novel task decomposition strategy will be formulated to decompose high
level task into a sequence
of sub
tasks or primitives, executable through multi
agent system. Throu
gh a centralized system, this
sequence can then be further planned into lower level primitives of agent. An interesting challenge will
be to construct the desired task behaviour through a decentralized system by self
organisation, through
the internal int
eraction among the agents and with the environment.

In environments where multiple robots are deployed, the proposed control system will be able to
communicate and negotiate between robots to allocate tasks to individual robots so as to ensure the
etion and the efficient flow of the high
priority tasks. The expected outcome of this work is core
theoretical research covering the above objectives and a prototype implementation that will be used in
empirical experiments to demonstrate this theory.

etable of Research

In the first year, the infrastructure essential to the project will be established utilizing FLOORBOTICS
8 floor
cleaning mobile robots. This includes hardware setup for several mobile robots and the
development of the localization a
nd map
building capabilities. This will involve the consolidation of
our research in the field and implementation of other established algorithms. Navigation capabilities
will also be quickly implemented from established algorithms. Once these infrastru
ctures are
established, research into multi
agent systems will be initiated by extending these fundamental
capabilities to handle multi
robot environment. At the same time, the formulation of task
decomposition will be further developed and established. V
arious task primitives in the form of self
contained modules containing low
level capabilities will be investigated and developed. An analysis of
various tasks to identify the unique primitives that make up the tasks will be performed and a
methodology to

generate a higher level task definition will be established.

The second year of the project will concentrate on establishing multi
agent algorithms for distributed
environments. Both centralized and decentralized algorithms will be evaluated and the op
timal strategy
will be established. Learning behaviours will be formulated to enable robust task and agent control to
cope with various changes and dynamics of task and the surroundings. Experiments will be conducted
to validate the effectiveness of the
behaviours through simulations and also through physical
implementation on the mobile robot research facilities. Furthermore, the experimental results will be
compared to the predictions from the theoretical models. Various human
robot interaction met
including multi
modal means, will be explored to obtain a set of robust and intuitive methodologies.

In the third year, methodologies for human
robot interaction, robot
robot interaction, task planning and
scheduling, and low
level primitives such

as path planning and mobile robot motion control algorithms
will be investigated and integrated to perform high
level cooperative task. Task decomposition,
planning and scheduling algorithms will be refined and improved with various application specific
requirements to obtain a truly generic set of task primitives. Validation will be carried out through both
simulation effort and real
time implementation on multiple mobile robots.



The results of this research are immediately applicable to various industries requiring efficient and
effective means of completing autonomous cleaning tasks. This could mean labour intensive
exhaustive tasks or tasks in hazardous areas.
The outc
omes of this project will help consolidate
Australia’s position in advanced research and innovative technologies and international developments
in the field of service robotics and advanced manufacturing and support systems.

The immediate application of
the result is geared towards autonomous cleaning tasks, whether it is
domestic (household) or even in the industry. Chemical spillage potentially stops a production line for
an extended period of time and causes considerable financial cost. Autonomous ro
bots, working
cooperatively to clean an area, could shorten the period of production shutdown as it is efficient and it
could be designed to withstand the hazardous material. This would also improve the safety of the
process and shorten the time normally
taken for extensive safety precautions when human cleaners are
involved. Other areas of application include delivery system (for example, within a hospital) and
building security system.

The resulting cooperative exploration and mapping methodologies co
uld also assist in many tasks such
as security patrol and in warehouses and the mining industry. An example in

shows a case of
autonomous exploration and mapping of abandoned mines. Utilising multiple small and lower
robots to perform the task also produces a more robust methodology and reduces the risk of failure
compared to using only one expensive robot to perform the job.

Extension of the algorithms into outdoor environment would provide an efficient and cost
method to explore, patrol, and monitor large areas. Cost
effective mass
produced agents in various
forms of robots could be dispersed autonomously across the vast landscape to form a sensor network,
perform exhaustive explorative mapping and moni
toring task, search
rescue, or even merely
providing continuously feedback on various types of information.


The outcomes of the research will be published in high quality refereed international journals and
presented at
refereed international conferences. Although the frequency of conference presentations
and journal publications heavily depends on the outcomes, we are determined to present at least one
refereed paper at international conferences in the first, second and

third years, and to publish one
journal paper in the second and third and fourth years. In total, three conference papers and three
journal papers are expected. A website will also be constructed and media articles will be prepared for
timely disseminatio
n of the outcomes.


Monash University

A/Prof. Shirinzadeh will manage the overall research activities. In addition, A/Prof. Shirinzadeh will be
responsible for supervision and training of the Research Associates (RAs), postgra
duate research
students, and Honours research students involved in the research program. He will also be responsible
for the research activities in the areas of task planning and navigation, hardware and software control
including experimental techniques s
uch as laser based measurement to verify motion strategies,
cooperative methodologies, and analysis and interpretation of theoretical and experimental results. The
CI will also be assisted by Research Associates (RAs) in the above tasks.

Dr. Ann Nicholso
n will provide expertise and leadership role in the area of intelligence to
accommodate uncertainties in the state ……

In addition, Dr Nicholson will provide assistance in supervision/co
supervision and training of the
Research Associates (RAs), postgraduat
e research students, and Honours research students involved in
the research program…..

Two full
time Research Associates (RAs) are being requested and will work on this project……..


Mr Murray McDonald is the Director of FloorBotics and will m
anage and lead the engineering, and
technical activities at FloorBotics. Mr McDonald will also supervise the engineers at FloorBotics and
organize system trials at sites such as industrial and household environments. Mr McDonald has been,
and will be, th
e driving force within this program to develop strategies for commercialization of the
developed technology. Mr McDonald has already generated, through extensive surveys and marketing
strategies, a great deal of interest among FloorBotics national and ove
rseas clients for FloorBotics
family of products/systems.



Anousaki, G.C.; Kyriakopoulos, K.J., Simultaneous localization and map building for mobile robot navigation,
Robotics & Automation Magazine
, Volume 6, Issue 3, Sept. 1999 Page(s)



Alami, R.; Fleury, S.; Herrb, M.; Ingrand, F.; Robert, F., Multi
robot cooperation in the MARTHA project,
Robotics & Automation Magazine
, Volume 5, Issue 1, March 1998 Page(s):36



Arkin, R.C., Integrating behavioral, perceptual, and
world knowledge in reactive navigation,
Robotics and
Autonomous Systems 6
, pp 105
122, June 1990.


Davison, A.J.; Murray, D.W., Simultaneous localization and map
building using active vision,
IEEE Transactions on
Pattern Analysis and Machine Intelligence
, V
olume 24, Issue 7, July 2002 Page(s):865



Di Marco, M.; Garulli, A.; Giannitrapani, A.; Vicino, A., Simultaneous localization and map building for a team of
cooperating robots: a set membership approach,
IEEE Transactions on Robotics and Automatio
, Volume 19, Issue
2, April 2003 Page(s):238



Dissanayake, M.W.M.G.; Newman, P.; Clark, S.; Durrant
Whyte, H.F.; Csorba, M., A solution to the simultaneous
localization and map building (SLAM) problem
, IEEE Transactions on Robotics and Automation
, Volume 17, Issue
3, June 2001 Page(s):229



Farinelli, A.; Iocchi, L.; Nardi, D., Multirobot systems: a classification focused on coordination,
IEEE Transactions
on Systems, Man and Cybernetics
, Part B, Volume 34, Issue 5, Oct. 2004 Page(s):201



Fong, T.; Thorpe, C.; Baur, C., Multi
robot remote driving with collaborative control,
IEEE Transactions on
Industrial Electronics
, Volume 50, Issue 4, Aug. 2003 Page(s):699



Ghidary, S.S.; Nakata, Y.; Saito, H.; Hattori, M.; Takamori,
T., Multi
modal human robot interaction for map
generation, Proceedings. 2001 IEEE/RSJ International Conference on Intelligent Robots and Systems, 2001, Volume
4, 29 Oct.
3 Nov. 2001 Page(s):2246



Guivant, J.E.; Nebot, E.M., Optimization of the sim
ultaneous localization and map
building algorithm for real
IEEE Transactions on Robotics and Automation
, Volume 17, Issue 3, June 2001 Page(s):242



Howard, A., Parker, L.E, and Sukhatme, G. S., The SDR Experience: Experiments
with a Large Scale
Heterogeneous Mobile Robot Team,
Proceedings of the Ninth International Symposium of Experimental Robotics
21 June 2004, Singapore, Page(s) 121



Kitano, H.; Asada, M.; Noda, I.; Matsubara, H., RoboCup: robot world cup
, IEEE Rob
otics & Automation Magazine
Volume 5, Issue 3, Sept. 1998 Page(s):30



Konolige, K. , Fox, D., Ortiz, C., Agno, A., Eriksen, M., Limketkai, B., Ko, J., Morisset, B., Schulz, D., Stewart, B.,
Vincent R., Centibots: Very Large Scale Distributed Robot
ic Teams,
Proceedings of the Ninth International
Symposium of Experimental Robotics
, 18
21 June 2004, Singapore, Page(s) 131



Lemoine, P.; Le Pape, C., Simulating actions and perception of autonomous mobile robots in a multi
agent indoor

Proceedings IROS '91. IEEE/RSJ International Workshop on Intelligent Robots and Systems '91
'Intelligence for Mechanical Systems, 3
5 Nov. 1991 Page(s):1479



Leonard, J.J., Durrant
Whyte, H.F., Simultaneous Map Building and Localization for an Aut
onomous Mobile Robot,
Proceedings of the 1991 IEEE/RSJ International Conference on Intelligent Robots and Systems '93
, IROS '91, 3
November 1991 Page(s):1442



Lima, P.U.; Saridis, G.N., Intelligent controllers as hierarchical stochastic automata,

IEEE Transactions on Systems,
Man and Cybernetics
, Part B, Volume 29, Issue 2, April 1999 Page(s):151



Logan, B.; Theodoropoulos, G., The distributed simulation of multiagent systems,
Proceedings of the IEEE
, Volume
89, Issue 2, Feb 2001 Page(



Mahadevan, S. and Connel, J., Automatic programming of behaviour
based robots using reinforcement learning.
Artificial Intelligence

3), Page(s): 311


McCarragher, B.J., Task primitives for the discrete event modeling and control of

DOF assembly tasks,
Transactions on Robotics and Automation
, Volume 12, Issue 2, April 1996 Page(s):280



McCarragher, B.J.; Asada, H., A Discrete Event Controller Using Petri Nets Applied To Assembly;
Proceedings of
the 1992 lEEE/RSJ Inter
national Conference on Intelligent Robots and Systems,

1992, Volume 3, July 7



Michelman, P.; Allen, P., Forming complex dextrous manipulations from task primitives,
Proceedings., 1994 IEEE
International Conference on Robotics

and Automation
, 1994, 8
13 May 1994 Page(s):3383



Morrow, J.D.; Khosla, P.K., Manipulation task primitives for composing robot skills,
Proceedings of IEEE
International Conference on Robotics and Automation
, 1997, Volume 4, 20
25 April 1997 Page(s



Murphy, R.R.; Lisetti, C.L.; Tardif, R.; Irish, L.; Gage, A., Emotion
based control of cooperating heterogeneous
mobile robots,
IEEE Transactions on Robotics and Automation
, Volume 18, Issue 5, Oct. 2002 Page(s):744



Nakauchi, Y.; K
awasugi, K.; Okada, T.; Yamasaki, N.; Anzai, Y., Human
robot interface architecture for distributed
Proceedings., IEEE International Workshop on Robot and Human Communication
, 1992, 1
3 Sept.
1992 Page(s):413



Parasuraman, S.; Ganapath
y, V.; Shirinzadeh, B.; Mobile robot navigation in an unstructured and unknown
environment using Al techniques,
Proceedings of the 7th Biennial Conference on Engineering Systems Design and
, 2004, 19
22 July 2004, Manchester, UK, p 295


man, S.; Ganapathy, V.; Shirinzadeh, B., Behavior based robot navigation: Resolving behavior conflicts
using fuzzy inference system,
Proceedings of the 7th Biennial Conference on Engineering Systems Design and

, v 2,19
22 July 2004, Manchest
er, UK, p 305


Parasuraman, S.; Ganapathy, V.; Shirinzadeh, B., Behaviour Coordination and Selection using Situation Context
Dependent Method for Behaviour based Robot Navigation using AI Techniques for Real World Environments,
Proceedings of the 5

ian Control Conference
, (IEEE and Institute of Engineers, Australia), 2004, p. 575



Saito, H.; Ishimura, K.; Hattori, M.; Takamori, T., Multi
modal human robot interaction for map generation, SICE
Proceedings of the 41st SICE Annual Conference
, Volume 5, 5
7 Aug. 2002 Page(s):2721



Shibata, T.; Fukuda, T., Coordinative behavior in evolutionary multi
robot system,
Proceedings of the 1993
IEEE/RSJ International Conference on Intelligent Robots and Systems '93, IROS '93
, Volume 1, 2
30 July 1993



Thrun, S.; Martin, C.; Yufeng Liu; Hahnel, D.; Emery
Montemerlo, R.; Chakrabarti, D.; Burgard, W., A real
maximization algorithm for acquiring multiplanar maps of indoor environments with mobile robots,
EE Transactions on Robotics and Automation
, Volume 20, Issue 3, June 2004 Page(s):433



Thrun, S.; Thayer, S.; Whittaker, W.; Baker, C.; Burgard, W.; Ferguson, D.; Hahnel, D.; Montemerlo, D.; Morris, A.;
Omohundro, Z.; Reverte, C.; Whittaker W, Au
tonomous exploration and mapping of abandoned mines,
Robotics & Automation Magazine
, Volume 11, Issue 4, Dec. 2004 Page(s):79



Tigli, J.Y.; Thomas, M.C., Use of multi agent systems for mobile robotics control
, IEEE International Conference on
ystems, Man, and Cybernetics
, 1994. 'Humans, Information and Technology'., Volume 1, 2
5 Oct. 1994



Timofeev, A.V.; Kolushev, F.A.; Bogdanov, A.A. Hybrid algorithms of multi
agent control of mobile robots,
International Joint Conference
on Neural Networks, 1999.(IJCNN '99)
, Volume 6, 10
16 July 1999 Page(s):4115



Zhang, O.; Yang, Y.; Li, Y., A multi
agent cooperative system of soccer robot,
Proceedings of the 4th World
Congress on Intelligent Control and Automation
, 2002, Volume 1
, 10
14 June 2002 Page(s):510