Training Simulation ofthe Manipulator Vehicle tEODor for

Explosive Ordnance Disposal (EOD) andImprovised Explosive Device Disposal (IEDD)

Wolfram SchoorEuropean Aeronautic Defence and Space Company (EADS)

Defence & Security - Military Air SystemsRechliner Strasse, 85077 Manching, Germany

fax: +49 (84 59) 81 88 80 997phone: + 49 (84 59) 81 80 997

email: wolfram.schoor@eads.com

Harald NikolisinEuropean Aeronautic Defence and Space Company (EADS)

Defence & Security - Military Air SystemsRechliner Strasse, 85077 Manching, Germany

fax: +49 (84 59) 81 80 531phone: + 49 (84 59) 81 81 076

email: harald.nikolisin@eads.com

Arne RadetzkyEuropean Aeronautic Defence and Space Company (EADS)

Defence & Security - Military Air SystemsRechliner Strasse, 85077 Manching, Germany

fax: + 49 (84 59) 81 80 938phone: + 49 (84 59) 81 80 755

email: arne.radetzky@eads.com

Abstract

The simulation of manipulator vehicles is a challenge in the field of EOD and IED disposal training. Thefocus in such vehicle simulations is to reproduce the behavior of the vehicle in terms of control andmovement and also to simulate interactions with the environment in order to prepare the user for realoperations.In this contribution an approach will be presented to ensure the quality of the implemented robot simu-lation for IED disposal training purposes.

Classification:I.2.9, I.6.1, J2

Keywords:Explosive Ordnance Disposal, Improvised Explosive Device, Evaluation, Robotics Simulation, PhysicsEngine, Interactive Environment

1

Introduction

The safety of life is one of the most important issues in the field of C-IED work. In the last decadesvarious tools and aids have been introduced to support the work of EOD/IEDD personnel. As shown in[Hes10] and [EM01] this kind of equipment can reduce the hazard to the EOD/IEDD personnel. One ofthe most powerful tools among those are the EOD/IEDD robots. They have been already used in Iraqto dispose of improvised explosive devices, nowadays a number of EOD robots are in deployment incountries throughout the world. The situation has changed in the last years, the number of IED attackshas drastically increased [Tod10]. As a tactical reaction, the application of IEDD robots also increased.To be best-possibly prepared for a hazard situation, EOD personnel must be trained in the usage ofEOD/IEDD robots.The robot hardware is very expensive and the number of devices is strongly limited. Often the deviceused for operations is also used for training purposes.The training of different operational situations canin reality not or only rarely be performed due to monetary related, time related reasons or availabilityissues. The application of virtual training can help to circumvent the practical limitations [JH09]. Thevirtual training can only be effective, if the virtual and the real training strongly correlate. That means,the obtained knowledge and skills can be applied to real life situations.To achieve this, a virtual training system respectively simulation environment must be intensively eval-uated and adapted. This work presents the evaluation of an IED disposal training simulation of thecommon EOD robot tEODor (Heavy Duty Robot class).

Related Work

The related work can be separated into three major parts. The first part contains a brief descriptionof the aim of EOD robots and an EOD robot classification. Furthermore, a set of distant EOD-relatedwork (not ground based tele-operated unmanned vehicles) will be listed. The second part contains aset of existing EOD/IEDD robot simulations, most with training focus. The third part contains a list ofEOD/IEDD related training software. A chapter summary concludes the results of the aforementionedparts.The interdependency of the topics EOD/IEDD robots, EOD/IEDD robot simulation and EOD/IEDD train-ing is visualized in Figure 1 as venn-chart (the intersection (gray) shows the concern of this paper).

EOD/IEDDrobots

EOD/IEDDrobot

simulation

EOD/IEDDtraining

Figure 1: Interdependency of related work

EOD/IEDD robots

EOD/IEDD robots are operator controlled robots (tele-operated) which support the EOD personnel inhazardous situations. These robots can also be used for ground surveillance, removal of hazardousmaterial (HAZMAT) or in chemical, biological, radiological, and nuclear (CBRN) scenarios.

2

How important these robots are, can be shown by the market volume of military ground robots, which ex-ceeded 0.8 billion US$ in 2009. In 2016 the figure of 9.7 billion US$ is anticipated (cf. [Eus10]). Differentclasses of EOD/IEDD robots exist based on their specific scope of duties. One common classification ispresented below:

• Super Heavy Duty• Heavy Duty• Light Duty• Super Light Duty

Figure 2 shows an example of a Light Duty Robot (a) and an example of a Heavy Duty Robot (b).

Figure 2: Examples of EOD robots: (a) Light Duty Robot (Image: Packbot from iRobot www.irobot.com);(b) Heavy Duty Robot (Image: Defender from Allen-Vanguard www.allenvanguard.com)

The key players1 in the field of professional EOD/IEDD in the Heavy Duty Robot class are: tEODor2,ANDROS Wolverine3, MARK3 (see Footnote 3, too), and Defender4. A comparison of the abilities andweaknesses of these EOD/IEDD robots is out of the scope of this paper.The majority of EOD robots belong to the category of teleoperated unmanned ground vehicles (UGV).Furthermore, unmanned underwater vehicles (UUV) or unmanned surface vehicles are used for EODlike [NLK+09]. Unmanned aerial vehicle (UAV) can support the C-IED work by protecting a convoy asshown in [Jon04]. For further readings on unmanned ground vehicles the authors refer to [SRBK08].

EOD/IEDD robot simulation

The simulation of robots is not restricted to training purposes. The initial application area for remoteassisted robot simulation has its roots in the field of software development. The computer simulationwas used in the early stage of robot software development to evaluate the similarity of real hardwarebehavior and simulated behavior. A systematic approach of EOD robot software verification and valida-tion can be found in [PBS07]. Such a systematic evaluation for the comparison of physical and graphicalperformance can also be used to evaluate the proposed training simulation software.The first generation of physics-based EOD training simulation software for the tEODor robot was devel-oped by the EADS one decade ago [EAD01]. Other developments in this area (with focus on trainingpurposes) are for example the EOD/IEDD robot simulation of the Packbot developed by [CL10] or thetEODor robot simulation provided by [vM09].A basic simulation for unmanned underwater vehicles applicable for EOD can be found in [HZAJZ08].In [CBM08] an environment for the integration of different robot simulations (UGV, UUV and UAV) foroperator training is presented.

1according to an internal study2Telerob GmbH http://www.telerob.de3Remotec http://www.is.northropgrumman.com4Allan Vanguard http://www.allenvanguard.com

3

EOD/IEDD training techniques

The training software for IED disposal can be separated in live training, computer assisted training,classroom instruction or a combination of the aforementioned (blended learning). IED disposal trainingin the form of life training is limited to the availability of the robot, (fake) IEDs and space. An appropriateexample of a blended EOD/IEDD learning approach can be found in [OB08]. It consists of a combinationof a live part-task trainer and a virtual reality system trainer [Ric08] combined with a game-based con-voy team-training (computer assisted and classroom instruction) to train an Army Clearance EngineerCompany.

Summary

The identified training gap between the required training of EOD personnel and the performed train-ing can be closed with an adequate robot simulation. As stated in the previous paragraph the robotsimulation should be integrated in the existing training infra structure. The simulation shall support theEOD/IEDD training to reduce the live training need for EOD personnel.

tEODor Simulation

This chapter roughly introduces the software design of the proposed tEODor robot simulation. In thefollowing, the necessary software components (internal and external) for the simulation are discussedand their basic tasks are explained.

Technical Design

The Figure 3 presents the overall software design for the EOD/IEDD robot simulation implementation.

Initialization Runtime Loop

RobotEngine plug-in

Fixed Setup Vortex Library

Vision Library

Vision Library

Input MapperOperator Console (serial link)Keyboard / Mouse Input

integrated component

Physic plug-in

Graphic plug-in

Deferred Shading

Virtual Input Device (VID)

plug-in component

Robot Simulation Application

Figure 3: Software design

The modularized design of the robot simulation application consists of the following modules: an inputmapper to interface with the operator console, a keyboard or virtual input device, a physic plug-in to thephysics engine, and a graphic plug-in to the graphics engine and a deferred shading component alsocommunicating with the graphics engine. In the following sections the parts of the graphical rendering,the physic rendering and the interface will be discussed in detail.

4

Graphical rendering

The graphical rendering is performed by the graphics engine also known as game engine or renderengine. It generates a rasterized image from a 3d model representation according to the renderingpipeline (see [AMHH08]). The representation may contain, beside the model data to render, meta-information about viewpoint, texture, lighting, and shading. Supported features of graphics engines maycontain inter alia:

• real-time rendering,• shading,• texture mapping / bump mapping,• shadows / soft shadows,• reflection / refraction / transparency / translucency,• indirect illumination,• depth of field / motion blur.

A variety of real-time graphics engines already exists on the market. The Vision Engine, Virtools, UnrealEngine, Shark3D, OpenSceneGraph, Onyx Engine, and OGRE are only a few common examples. Adetailed consideration on graphics engines is out of the scope of this work. In [JW09] a practical analysisof selected graphics engines was performed.The implementation of the graphics rendering software part should benefit from an existing and approvedsoftware solution in order to circumvent “the reinvention of the wheel”. The integration of third partylibraries like A.I., sound, physics, etc., must be supported, too. Due to already existing interfaces tothe above mentioned third party libraries, high-end graphics, a professional editor, and a direct productsupport the Vision Engine became the engine for the graphical output.

Physic rendering

The term physics engine refers in general to a part of a computer program with the aim of calculatingphysical processes in real-time, for example the dynamic behavior of a robot. The alternative wouldbe high precision simulation software which includes CFD (Computation Fluid Dynamic) or FEM (FiniteElement Method) which deliver very accurate results at the expense of computational time.Beside the standard calculation of physics on single or multi CPUs (possible with double precision arith-metic), special hardware units have been introduced for physics processing (PPU) by Ageia with nosignificant impact. By using the massively parallel computing capabilities of the GPUs (graphics pro-cessing unit) for physics calculation based on “textures” as shown in [Har07], opened the way for thenowadays common approach of General-Purpose computation on GPU (GPGPU) (cf. [OLG+07]). Thebenefit of new GPU hardware development (according to the IEEE 754− 2008 standard) allows the cal-culation of physics in double precision arithmetic on modern GPUs.A wide range of physics engines exists on the market, either OpenSource engines like Bullet, IBDS,Tokamak, and Open Dynamics Engine (ODE) or proprietary engines like Intel’s Havok, nVIDIA’s PhysX(freeware), or Cm-Lab’s vortex, to name a few. An evaluation of open source physics engines can befound in [SR06]. The domain of these engines is rigid body dynamics. If supported, the handling of softbody dynamics and fluid dynamics is mostly performed as a strong simplification. For further readingson physics algorithms the authors refer to [Bou01] or [Eri04].An example of real-time CFD calculation can be found in [MCG03], whereas the simulation of soft-bodydynamics is described in [CTA+08].In [GTG08] different physics engines were evaluated with focus to stable contact simulation behavior,which is a prerequisite for grasping. Only nVIDIA’s PhysX and Cm-Lab’s vortex showed stable and plau-sible results in this study.The authors of this paper also evaluated both of these engines. Problems with the reaction, generatedby the PhysX constraints and a much higher accuracy and robustness of the vortex engine, caused thatvortex was used for the final implementation. Vortex was integrated to Trinigy’s Vision Engine to gainaccess to its scene graph. The basic steps to setup the physic simulation were:

• defining materials and contact properties,• defining the rigid body entities with collision geometry,mass, center of gravity,• defining constraint elements between the rigid body entities of the robot,

5

• defining kinematic constraints to represent real world limits,• defining actuators for moving the robot parts, and• establishing the drive system for the robot.

Every step of the simulation, the vortex core solver calculates the results and delivers the positionupdates to the graphics engine. Additionally, the main loop subscribes to the notification of contact andconstraint information to adjust kinematic conditions and deliver various interface components.The use of coulomb friction in the simulation invokes the Linear Complementarity Solver [Fea87] whichiteration parameters are controllable. Generally, “tweaking” with these parameters, as well as dampingand compliance settings is needed to maintain an acceptable degree of performance. For the samereason the amount of contact possibilities should be minimized. The result of these efforts is a reliableand an accurate physical behavior of the robot simulation.

Interface

The application is a modulated object-orientated program written in C++. The core application has threeplug-ins:

• plug-in for the graphical engine,• plug-in for the physics engine,• plug-in for describing the robot parts.

The graphical engine plug-in is the connection to Trinigy’s Vision Framework, whereas the physics engineconnects to the Cm-Labs vortex library. The Robot Engine itself can describe various robots, currentlythe telerob tEODor is implemented and telerob telemax is under development.Furthermore, additional abstraction layers exists in the application, like the input device layer. Thiscomponent handles the various control devices, from a simple computer keyboard to the original robotcontrolling devices, like the operator console of tEODor connection via a serial link.Finally, a variant management system allows a flexible configuration of the robot containing differenttools and parts.

Test Methods for Objective Verification

According to different approaches such as [PBS07] or [XCJT06] a standardized procedure helps tobenchmark the quality of a simulation. For the evaluation, relevant objects of a physically existing trackwere virtually rebuilt as a reference environment. The major purpose of this objective verification is theevaluation of the EOD robot’s functionality and its intrinsic behavior. A subset of the evaluated qualitysubjects are listed below. Figure 4 shows the real track and its virtual equivalent.

Figure 4: Detail of the evaluation track: (a) real track; (b) virtual track.

6

Kinematic model

The velocity of the vehicle depends mainly on two factors, the physical engine power and the interactionof the robot with the surface on which it is operating. The engine power was specified by the manufac-turer and the contact parameters were adjusted according to the materials. The driving characteristic ofthe robot is listed in Table 1.

Robot Speed Real (sec) Virtual (sec)driving 6.3 6.3turning 4.2 4.4

Table 1: Driving characteristics

Another pool of movements is related to the manipulator of the robot. This includes the upper arm andlower arm movements, telescope spindle, and the movement of the gripper including the turn and tilt ofthe wrist. Table 2 contains the results of the manipulator movement evaluation.

Function Real (sec) Virtual (sec)rotational speed of manipulator 40 42tilt speed lower arm 16 18tilt speed upper arm 16 18speed of telescope spindle 22 24wrist turn 9 10wrist tilt 24 26closing gripper 20 19opening gripper 30 28auto. motion “initial state” 23 25packing position 19 21

Table 2: Manipulator characteristics

Step test by using stairs

The step test is conceived for the analysis of the robot’s climbing capabilities. It is very important for thebehavior of the robot to achieve comparable results between the real and virtual robot’s performance.Figure 5 illustrates the quality between a real step and a virtual step on a stair. Table 3 shows thepractical results for inclinations of 32, 38, and 45 degrees.

Figure 5: Example of the step test verification: (a) real step; (b) virtual step.

Degree Real Virtual32 passed passed38 passed passed45 passed passed

Table 3: Step test by using stairs with different inclinations.

7

Tilt test

One important indicator for a correct weight distribution in the virtual robot design is the tilt test. Precon-dition for a valid virtual robot design is the correct mass distribution for each model part respective to themanufacturers’ specification. Given a predefined arm position, the beakover point must be nearly iden-tical between real and virtual model. The result of the breakover point for one predefined arm position(packing position) is shown in Figure 6. Table 4 lists the results in numbers.

Figure 6: Tilt test - determination of the breakover point: a) real model; b) virtual model

Pose Real (degree) Virtual (degree)packing position 64 63craned arm (telescope in) 56 58craned arm (telescope out) 54 54auto. motion “initial state” (turned) 52 52

Table 4: Results of the tilt test.

Subjective Verification

Due to the unavailability of resources and limitations in the variability of operational situation in the train-ing (see Chapter Introduction) the operational verification was not possible for all simulation components(especially external). Therefore, specialists of the military personnel and also from the manufacture’sside evaluated the plausibility of events, behavior and appearance.

Explosives

The shape and form of a real IED can vary widely but consists in general of the following five parts:

1. an initiation system or fuze,2. explosive fill,3. a detonator,4. a power supply for the detonator, and5. a container.

The simulation system contains a number of predefined IED shapes.

• suitcase,• package,• metal box,• pipe bomb,• fire extinguisher bomb.

8

Figure 7: Shooting an IED at the rifle range (external camera view): (a) Hitting the IED; (b) Explosion ofthe IED.

All of the IED shapes can be combined with different types of detonation mechanisms: Detonation bytime, Detonation by acceleration and Detonation by inclination (see Figure 7).The IEDs were subjectively evaluated according to their tangibility, physical behavior, sensor behaviorand explosion effect. The IED features mentioned previously all showed plausible results.

Weather effects and dynamic effects

The following weather effects can be selected:

• rain with varying intensity,• snow with varying intensity,• sand storm,• volumetric fog,• dust,• Continuous time of day (CTOD); dawn, daylight, dusk, and night.

In Figure 8 two possible weather conditions (dust and snow) as well as a fire effect are presented. Alldescribed visual effects are equivalent to their real phenomena.

Figure 8: Effects in different data bases (external camera view): (a) Sanddust and fire effect; (b) Snowfall.

9

Weapons and tools

The listed weapons & tools are included in the presented simulation. The usage of the virtual weaponand its escapement are plausible and show good looking visual effects. The gripper performance offerscorrect physical simulation results due to the applied vortex gripping algorithms.

• shotgun (see Figure 9),• gripper,• camera mast (telescopic),• mast camera and elbow camera,• target marking LASER,• telemeter (distance measurement) [prototype],• Unattended Luggage Inspection System (ULIS) [ES09] [prototype].

Figure 9: Detail view of shotgun fire.

Virtual components

Virtual components include all the computer models needed to simulate the surroundings (environment)of the robot simulation. In the Figures 4 (b), 8 (a) and (b), and 12 the close to reality representation ofthe 3d models can be seen.The physical model behavior of the 3d computer models of flexible objects (like boxes, IEDs, doors,stones) must be designed in addition, according to [CL09]. A physical level of detail can be used, tomeet the training needs and keep the simulation real-time. Static objects (like houses or walls) are partof a static mesh which is used for the collision detection with flexible objects or object parts. In Figure 10the geometrical representation and the physical object representation are illustrated for two differentobjects.

Figure 10: Geometrical and physical model representation: a) suitcase; b) Traffic cone.

10

Lighting

A correct lighting of the virtual scene is necessary to provide the EOD personnel with the same percep-tion. The lighting must be differentiated between the environmental lighting and the robot lighting. InFigure 11 the real lighting and the virtual lighting (both with the robot’s lights) are shown.

Figure 11: Robot lights: a) lights of the real robot; b) lights of the virtual robot.

The cone size of the lights, the lighting intensity and the intensity gradient show comparable results. Thegripper light interacts with the gripper and creates self-shadowing (see Shadows)

Shadows

Shadows are one of the most powerful key indicators for depth perception [SBS+10], especially if nostereo vision is present. To know exactly the relative robot’s position (in particular the gripper) to the IEDis very important to guarantee the security of the current IEDD operation. Hitting or touching an IED byincident (e.g. due to a wrong depth estimation) can trigger the fuze mechanism of the explosive. Thesimulation supports real-time hard shadows as well as soft shadows. An important implementation detailis the availability of self-shadowing. That means one object can create a shadow on itself. Figure 12illustrates examples of realistic shadow generation in the proposed virtual simulation environment. Fur-ther information about (soft) shadow generation techniques can be found in [AMA02].

Figure 12: Shadows in the simulation environment: a) dynamic soft shadows in combination with bumpmapping; b) shadow generation based on multiple light sources and varying intensities, e.g. at theguardrail.

Conclusion

In this work the evaluation of a Heavy Duty Class robot - the tEODor - was presented. It could be shownthat the simulation behavior and the real performance of the robot have a very strong correlation. Theresults of the evaluation (objective and subjective) are summarized in Table 5.

11

Verification Type Object of Investigation Status

Objective

Kinematic model passedStair test passedTilt test passed

Subjective

Explosives passedWeather effects / Dynamic effects passedWeapons and tools passedVirtual components passedLighting passedShadows passed

Table 5: Summary of the evaluation.

Future Work

The results presented here are part of an evaluation performed on the robot simulation software. Amanufacture certification of the tEODor simulation is currently in progress.The implementation of other EOD robots (telemax robot, see Figure 13 ) is the strategical future goal toprovide EOD personnel with a wider range of training possibilities. This allows a more flexible approachfor IED disposal fitting the requirements for the mission to accomplish.

Figure 13: telemax virtual robot model for IEDD training purposes.

For the tEODor, improvements in the implementation of sound, material coefficients, lens distortion andadditional equipment is planned. The robot simulation was integrated into an IEDD scenario trainingdemonstrator. This demonstrator combines the training of handling a hazardous situation and trainingprofessional skills respectively specialized knowledge [EAD10].

Acknowledgment

The authors would like to thank the telerob GmbH (manufacturer of the tEODor EOD robot) for theexcellent support with all necessary information.

About the Authors

Wolfram Schoor was born on September 20th, 1977 in Anklam, Germany. After finishing his bachelorand diploma in the field of computational visualistics at the Otto-von-Guericke University in 2003 and2004, he worked as researcher in the Fraunhofer Institute for Factory Operation and Automation (IFF) inMagdeburg in different scientific projects with focus on simulation and visualization. In 2007 he becamethe leader of the Biological Visualization Group in the department Virtual Prototyping.In 2009 Mr. Schoor changed his position to the department Engineering Training Systems at the com-pany EADS Defence & Security. Here, he is responsible for the robot simulation projects.His present research interests are related to visualization, virtual/augmented reality and simulation.

12

Harald Nikolisin finished 1998 his study of civil engineering with diploma at the University of Stuttgart.His specializations in informatics and dynamics brought him to SOFiSTiK AG near Munich, where heworked as a software developer.He worked mainly on the FEM dynamic programs and 3D visualization of FEM results and maintains theUNIX Versions of the whole software module range.Early 2008 he moved to EADS Defence & Security to the department of Simulation Combat Air Systems.Besides working with Linux Realtime systems he is responsible for visualization techniques as well asphysic rigid body simulation. Since 2009 he is the chief developer of the robot simulation.

Dr. Arne Radetzky (PhD in Computer Sciences) has worked for 14 years in the area of ComputerGraphics, Virtual Reality and Simulation. He invented the neuro-fuzzy deformation and successfullytested this technique in some of the first Surgical Simulators in the world and in military simulators forthe training of Special Forces.He is author of more than 40 peer-reviewed scientific publications about Virtual Reality and Simulation.He gave lectures on Virtual Reality and Simulation in several European Universities, e.g. as Professorat the Imperial College of Science, Technology and Medicine, London. He was speaker at more than 20international conferences around the world.Since 2006 he is responsible for the development of training media including simulators for Military andGovernmental agencies as head of the department Engineering Training Systems at EADS Defence &Security.

References[AMA02] T. Akenine-Möller and U. Assarsson. Approximate soft shadows on arbitrary surfaces us-

ing penumbra wedges. In EGRW ’02: Proceedings of the 13th Eurographics workshopon Rendering, pages 297–306, Aire-la-Ville, Switzerland, Switzerland, 2002. EurographicsAssociation.

[AMHH08] T. Akenine-Möller, E. Haines, and N. Hoffman. Real-Time Rendering 3rd Edition. A. K.Peters, Ltd., Natick, MA, USA, 2008.

[Bou01] D. M. Bourg. Physics for Game Developers. O’Reilly Media, Inc., 2001.

[CBM08] J. Craighead, J. Burke, and R. Murphy. IEEE/RSJ 2008 International Conference on In-telligent Robots and Systems: Workshop on robot simulators: available software, scientificapplications and future trends. In Using the Unity Game Engine to Develop SARGE: A CaseStudy, Nice, France, September 22 2008.

[CL09] CM-Labs. Vx: Vortex C++ API - Developer Guide. Technical report, CM-Labs Simulation,Inc., 2009. Version 4.1.

[CL10] Cm-Labs. Vortex. http://www.vxsim.com, 2010.

[CTA+08] O. Comas, Z. A. Taylor, J. Allard, S. Ourselin, S. Cotin, and J. Passenger. Efficient NonlinearFEM for Soft Tissue Modelling and Its GPU Implementation within the Open Source Frame-work SOFA. In ISBMS ’08: Proceedings of the 4th international symposium on BiomedicalSimulation, pages 28–39, Berlin, Heidelberg, 2008. Springer-Verlag.

[EAD01] EADS Defence & Security. Training EOD - Explosive Ordnance Disposal: Simulator-Manipulationsfahrzeug RED - Robot Explosive Disposal. Product Flyer, 2001.

[EAD10] EADS Defence & Security. IEDD Scenario Training. Product Flyer, 2010.

[EM01] Defense Media Activity Emerging Media. The bomb suit’s always been a matter of trust.http://www.defence.gov, 2001.

[Eri04] C. Ericson. Real-Time Collision Detection (The Morgan Kaufmann Series in Interactive 3DTechnology). Morgan Kaufmann Publishers Inc., San Francisco, CA, USA, 2004.

[ES09] EADS-Sodern. Unattended luggage inspection system for homeland security and defense.Product Flyer, 2009.

[Eus10] S. Eustis. Military robots and unmanned vehicles market shares strategies, and forecasts,worldwide. Technical report, WinterGreen Research Inc., 2010.

13

[Fea87] R. Featherstone. Robot dynamics algorithms. Kluwer Academic Publishers, 1987.

[GTG08] M. Gienger, M. Toussaint, and C. Goerick. Task maps in humanoid robot manipulation. InIROS, pages 2758–2764, 2008.

[Har07] T. Harada. GPU Gems 3, chapter Real-Time Rigid Body Simulation on GPUs. Addison-Wesley, 2007.

[Hes10] A. Hesse. Bodengebundene unbemannte Systeme. Strategie und Technik, 4:14–17, 2010.

[HZAJZ08] M. S. M. Hamzah, M. Zakaria, M. F. I. Abd Jalil, and K. Z. Zamli. 3d virtual simulationsoftware for underwater application. In 2nd International Conference Underwater SystemTechnology, 2008.

[JH09] K. Jenewein and D. Hundt. Wahrnehmung und Lernen in virtueller Realität - Psychologis-che Korrelate und exemplarisches Forschungsdesign. Technical report, Otto-von-GuerickeUniversity, 2009. IBBP-Arbeitsbericht Nr. 67.

[Jon04] H. Jones. Mini-UAVs for Convoy Protection. In Unmanned Systems, volume 22, pages13–16, 2004.

[JW09] N. Johansson and F. Williamsson. Human machine interface visualization enhancement ofan abb quality control system. Master’s thesis, Umeå University, 2009.

[MCG03] M. Müller, D. Charypar, and M. Gross. Particle-based fluid simulation for interactive appli-cations. In SCA ’03: Proceedings of the 2003 ACM SIGGRAPH/Eurographics symposiumon Computer animation, pages 154–159, Aire-la-Ville, Switzerland, Switzerland, 2003. Eu-rographics Association.

[NLK+09] H. G. Nguyen, R. Laird, G. Kogut, J. Andrews, B. Fletcher, T. Webber, R. Arrieta, and H. R.Everett. Land, Sea, and Air Unmanned Systems Research and Development at SPAWARSystems Center Pacific. SPIE, 2009.

[OB08] M. O’Bea and J. Beacham. How do You Train When Your Equipment is 7000 Miles Away?In ITEC 08 Proceedings, 2008.

[OLG+07] J. D. Owens, D. Luebke, N. Govindaraju, M. Harris, J. Krüger, A. E. Lefohn, and T. J. Purcell.A survey of general-purpose computation on graphics hardware. Computer Graphics Forum,26(1):80–113, 2007.

[PBS07] C. Pepper, S. Balakirsky, and C. Scrapper. Robot simulation physics validation. In PerMIS’07: Proceedings of the 2007 Workshop on Performance Metrics for Intelligent Systems,pages 97–104, New York, NY, USA, 2007. ACM.

[Ric08] A. Rick. Left of Boom. MS&T Magazine, 1:7–10, 2008. Virtual Route Clearance Trainer(VRCT) by Raydon Corp.

[SBS+10] W. Schoor, F. Bollenbeck, T. Seidl, D. Weier, W. Weschke, B. Preim, U. Seiffert, andR. Mecke. VR Based Visualization and Exploration of Plant Biological Data. Journal ofVirtual Reality and Broadcasting, 6(8), 2010. VRIC 2009 Special Issue.

[SR06] A. Seugling and M. Rölin. Evaluation of Physics Engines and Implementation of a PhysicsModule in a 3d-Authoring Tool. Master’s thesis, Umeå University, 2006.

[SRBK08] D. P. Sellers, A. J. Ramsbotham, H. Bertrand, and N. Karvonides. International Assessmentof Unmanned Ground Vehicles. Technical report, Institute for Defense Analyses (IDA), 2008.

[Tod10] B. Todd. Interview: Simulating roadside bombs. CNN - Video, 2010. http://edition.cnn.com/video/?/video/tech/2010/03/29/todd.countering.ieds.cnn.

[vM09] A. von Michel. Aus der Gefahrenzone: Realitätsnahes Training in virtuellen Szenar-ien. Virtual Reality Magazin: Visualisierung, Simulation, Interaktion, 2:26–27, 2009.www.virtual-reality-magazin.de.

[XCJT06] L. Xuewen, M. Cai, L. Jianhong, and W. Tianmiao. Research on Simulation and TrainingSystem for EOD Robots. In 2006 IEEE International Conference on Industrial Informatics,pages 810–814, 2006.

14