mixed robot formations afases

7/29/2019 Mixed Robot Formations AFASES

1/7

HENRI COANDA GERMANY GENERAL M.R. STEFANIKAIR FORCE ACADEMY ARMED FORCES ACADEMY

ROMANIA SLOVAK REPUBLIC

INTERNATIONAL CONFERENCE of SCIENTIFIC PAPERAFASES 2011

Brasov, 26-28 May 2011

A FRAMEWORK FOR CREATING MIXED ROBOT FORMATIONSWITH PHYSICAL AND VIRTUAL AGENTS

Ioan SUSNEA, Grigore VASILIU

University Dunarea de Jos of Galati, Romania

Abstract: This paper proposes a solution for creating mixed formations with virtual and physical agentsand describes a framework for studying stigmergy in such formations. The central element of the solutionis the concept of virtual pheromones, defined as engrams created by the agents not in the environment,but in a map of the environment, stored by a pheromone server. This map acts like a shared memory areafor all the agents, embedding information usable for defining paths and actions, and is dynamicallyupdated by the server, which is in permanent communication with the agents, through a radio link. Amodel of the virtual pheromone is proposed, describing the spatial diffusion, and evaporation. Possibleuses of this concept for controlling the real-time motion of the agents are explored..Keywords: Virtual pheromones, pheromone server, spatial diffusion, controlling the real-time motion.

1. INTRODUCTION

Afterwards, a great number of scientific papersproposed various methods for creatingartificial pheromones. Some researcherspropose solutions based on spreadingchemicals in the environment, just like ants do.(Purnamadjaja 2007, and Genovese 1992).Others (Payton 2005) use short-range infraredtransceivers to relay messages between mobile

robots, while others (Mamei 2005, Susnea2008) propose the use of RFID tags, deployedin the environment, to store some datastructures, interpreted asdigital pheromones.

The term virtual pheromone was mainlyused in connection with software agents(Szumel 2006).

In all of the above implementations, artificialpheromones are psysical, chemical or

informational entities distributed in theenvironment, thus having the benefit of theintrinsec robustness of any decentralized

multi-agent system.This paper presents a

centralized approach on using artificialpheromones, wherein virtual pheromones areengrams created by the agents not in theenvironment, but in a map of the environment,located in a pheromone server.

In this approach, the agents use their ownlocalization system to periodically report theirposition to the pheromone server, via a radiocommunication link. When the pheromoneserver receives a data packet containing thecurrent position of an agent, it locates theagent on the internal map, then computes thepheromone concentrations for that particularposition, and sends back to the client aresponse packet containing this data.

Thereafter, the agent acts as if it had its owndifferential pheromone sensors, and adjusts itsposition so that it gets as close as possible tothe pheromone trail. The system can operatewith fixed, predefined paths embedded in the

pheromone map, or, when multiple agents areinvolved, it can modify the pheromoneconcentrations as if the agents would leave

710


2/7

pheromone trails on their way, just like realinsects do. In this last case, the pheromonepaths stored by the server dynamically changeas agents move through the environment,

creating a realistic emulation of a naturalswarm.

Beyond this introduction, this paper isstructured as follows:

Section 2 proposes a model of the virtualpheromones

Section 3 contains a description of anexperiment where physical agents follow avirtual leader by means of a pheromoneserver.

Section 4 presents some experimentalresults, and

Section 5 is reserved for discussion andfuture work.

2. A MODEL OF THE VIRTUALPHEROMONES

Natural pheromones are chemical substancesreleased in the environment by some insectsand other animals, in order to influence thebehavior, and sometimes even the physiologyof other members of the same species.

Ant foraging is the most common example ofpheromone-based interaction. When an antfinds a food source, it starts spreadingpheromone on its way back to the nest, leavinga trail that indicates the path to the food to the

other ants. Every ant that senses thepheromone trail tends to follow the existingpath and reinforces the pheromone trail byspreading additional pheromone.

On the other hand, the pheromone is subject toevaporation, and, when the food source isexhausted, in the absence of reinforcement, thetrail disappears. This indirect coordinationbetween agents by means of modifying theenvironment by an action, which stimulatessimilar subsequent actions, in a positivefeedback process, is called stigmergy (Grass1959).

Any model of the natural pheromones shouldaddress at least the following aspects:

Spatial diffusion.Pheromone diffusion gradients providevaluable navigational information and alsoencode useful information about obstacles thatobstruct pheromone propagation. Normally,insects sense the pheromone by means of twomovable antennas located on the sides of thehead. This allows the insect to sense the spatialgradients of the distribution of the pheromone.

Superposition of the effects of multiplepheromone sources.

The overall intensity of the pheromone, sensedat a given point, is a result of thesuperposition of the effects of multiplepheromone sources. Figure 1 illustrates thismechanism.

Evaporation.The intensity of the pheromone effect decreasewith time. This process reduces obsolete orirrelevant information, and also provides thecolony with a mechanism to find the shortestpath to the food. If two or more paths ofdifferent lengths are available between the nestand the food source, longer paths require moretime for the ants to reach their target. Theprocess of evaporation reduces the intensity ofthe pheromone on longer paths, and thereforemore ants tend to choose the shorter path.Eventually, the shortest path is used by mostof the ants, and longer paths disappear. Seefigure 2 for an illustration of this process, as

explained in the so-called double bridgeexperiment (Deneubourg 1990).

Fig. 1 An illustration of the spatial diffusion andsuperposition.

711


3/7





Fig. 2 The double bridge experiment

The spatial diffusion can be modeled with (1):

=

0

1)( 0

xp

xp


4/7

formations including physical and virtualagents.

The simplest cooperative behavior followthe leader - was selected, with the difference

that the leader was a simulated, virtual robot,(simulated with MobileSim) while thefollower was a physical robot, namely therobot Pioneer3-DX. Both the robot and thesimulator MobileSim are manufactured byMobileRobots Inc. (MobileRobots, 2000).

The pheromone server was implemented witha desktop computer, connected through a IEEE802.11g WLAN with the agents.

3.2Experimental Setup

The structure of the equipment used in theexperiment is presented in figure 3.

Fig. 3 The Experimental Setup

The motion of the simulated robot wasmanually controlled. A low cost/low powermicrocontroller unit (MCU) implemented afuzzy controller for path following, bygenerating reference values for the speeds vL,vR of the left and right wheels of thedifferential drive robot, based on thepheromone intensities PL, PR, reported by theserver. See (Susnea 2008b) for details on thefuzzy controller.

3.3Notes on the Actual Implementation

The environment map used was a simple 2Dgrid map whose nodes were linked with a datastructure containing information aboutpheromone source (if any) and a time stamp.Each cell of the map corresponds with a squarearea in the real world (see figure 4).

A special software application was written togenerate and manage such maps. See figure 5for a snapshot of the GUI of the map editor.

Fig. 4 The grid map embedding information on thepheromone distribution

The robots sensitivity to pheromones wasassumed to be directional, so that only a 180degrees circular sector ahead of the robot isvisible for the pheromone sensing antennas.

In the first phase of the experiment, the robotswere instructed to follow a pre-definedpheromone trail. This was used to tune theFLC for path following and to test thecommunication functions.

In the second phase of the experiment, thepheromone trail was generated by recordingsuccessive positions of the simulated robot,and the physical robot was instructed to followthe pheromone trail created by the virtualleader.

In both cases, the pheromone trace wasassumed to be time-invariant the evaporationwas not taken into consideration.

The environment was assumed to be ahorizontal plane with no obstacles. Theodometric system of the robot was used forposition estimation.

Fig. 5 Snapshot of the GUI of the map editor

4. EXPERIMENTAL RESULTS

Figures 6 and 7 present the actual path of asimulated robot, recorded with MobileSim,

713


5/7





superposed with the corresponding pheromonedistribution map for pre-defined paths.

The path used for recording the evolution ofthe robot in figure 8 was generated by a virtualleader robot, manually controlled.

In all cases, the follower robot was controlledby a fuzzy logic controller implemented on alow cost microcontroller.

Fig. 6 Recorded path vs pheromone distribution

Fig. 7 Recorded path vs. pheromone distribution

Fig. 8 Recorded path vs. pheromone distribution

5. DISCUSSION AND FUTURE

RESEARCH

This centralist approach has been criticisedand a-priori rejected (Parunak 2002) for lack

714


6/7

of robustness in case the pheromone serverfails. However, this limitation can be easilyeliminated by using a second backup server,which executes all the operations of the mainserver, except responding to the queries of therobot clients. If the main server fails tocommunicate for a specified time-out delay,the backup server automatically assumes thecommunication tasks. There are, instead,several important advantages of this method:

- It distributes the computational loadbetween the server and the agents, thusallowing very simple control structures forthe agents. This results in drastic costreduction thereof.

- Convenient pheromone distribution maps,obtained, for example, in an Ant ColonyOptimization process, can be saved forlater use. Virtual agents can be used in theoptimization process, and the results areimmediately available for use withphysical robots.

- Multiple pheromone types can be defined(e.g. stay away from this area, find and

pick a load, drop the load here etc.).- Virtual leader-robots, with high speed and

maneuerability can be sent in real-time toinfluence and guide formations of physicalrobots.

- This method is applicable to control anytype of military, civil, and industrialservice robots. Existing robots can beeasily modified to suit this control method.

One major drawback of the proposed methodis that it relies on the agents localizationsystem. For outdoor applications the GPSlocators can give satisfactory results, butindoors a localization system better than deadreckoning must be implemented. Furtherresearch is needed in the following directions:

- Define a method to embed informationabout the obstacles in the pheromonedistribution maps, and describe howobstacles influence pheromone diffusion.

- Experiment with other types ofpheromones.

- Identify other possible applications of themethod.

REFERENCES

[1] Beni, G., Wang, J. (1989) Swarmintelligence in cellular robotic systems,Proceedings of NATO Advanced Workshopon Robots and Biological Systems,

Tuscany, Italy, , pp:2630[2]Bonabeau. E., Dorigo M, and G. Theraulaz,

(1999) Swarm intelligence: from natural toartificial systems. , Oxford UniversityPress, 2004

[3] Deneubourg, J.-L., Aron, S., Goss, S. andPasteels, J. M. (1990), The self-organizing

exploratory pattern of the Argentine ant,J .Insect Behav. 32, pp:159- 168.

[4] Dorigo M et al. (1996) The ant system:optimization by a colony of cooperatingagents. IEEE Trans Syst Man Cybernet26(1), , pp:2941

[5] Karlson, P., Lscher, M. (1983)Pheromones: a new term for a class ofbiologically active substances. Nature183,pp. 55-56

[6] Grass P.P. (1959) La Reconstruction du

nid et les coordinations inter-individuelleschez Bellicositermes Natalensis etCubitermes sp. La theorie de la stigmergie:essai dinterpretation du comportement destermites constructeurs. Insectes Sociaux, 6

[7] Genovese, V., Dario, P.,Magni, R., &Odetti, L. (1992), Self organizing behaviorand swarm intelligence in a pack of mobileminiature robots in search of pollutants. InProceedings of the IEEE/RSJ internationalconference on intelligent robots andsystems, Raleigh, NC, 1992, pp. 15751582

[8]Mamei M., F. Zambonelli, (2005) Physicaldeployment of digital pheromones throughRFID technology, Swarm IntelligenceSymposium, 2005. SIS 2005.

[9] Parunak V.D., Purcell M., OConnell R.(2002), Digital pheromones for autonomouscoordination of swarming UAVs,.Available online:

http://www.newvectors.net/staff/parunakv/AIAA-2002-3446.pdf[Online][10] Payton D., R. Estkowski, M. Howard,

715


7/7





(2005) Pheromone robotics and the logic ofvirtual pheromones, in Swarm Robotics,Springer, 2005, pp: 45-57

[11] Purnamadjaja A.H, Russell R. A, (2007)Guiding robots behaviors using pheromoneommunication J ournal of AutonomousRobots, Vol. 23 (2), Kluwer AcademicPublishers Hingham, MA, USA, 2007, pp:

113 130[12] Susnea I., G. Vasiliu and A. Filipescu,

(2008a) RFID Digital Pheromones forGenerating Stigmergic Behaviour to

Autonomous Mobile Robots, Proceedingsof the 4th WSEAS/IASME InternationalConference on dynamical systems andcontrol (CONTROL'08), Corfu, Greece,October 26-28, 2008, ISSN: 1790-2769.

[13] Susnea, I.; Filipescu, A.; Vasiliu, G.;Filipescu, S. (2008b), Path following, real-time, embedded fuzzy control of a mobile

platform wheeled mobile robot,Automationand Logistics, 2008. ICAL 2008. IEEEInternational Conference Volume 1 , Issue, 3 Sept. 2008, pp:268 272

716

mixed robot formations afases

Documents