University of PennsylvaniaScholarlyCommons

Center for Human Modeling and Simulation Department of Computer & Information Science

3-1-1995

MediSim: Simulated Medical Corpsmen andCasualties for Medical Forces Planning andTrainingNorman I. BadlerUniversity of Pennsylvania, badler@seas.upenn.edu

John R. ClarkeMedical College of Pennsylvania

Michael J. HollickUniversity of Pennsylvania

Evangelos KokkevisUniversity of Pennsylvania

Dimitris MetaxasUniversity of Pennsylvania

See next page for additional authors

Copyright 1995 IEEE. Reprinted from Proceedings of the National Forum on Military Telemedicine On-Line Today, 1995, March 1995, pages 21-28.Publisher URL: http://dx.doi.org/10.1109/MTOL.1995.504523

This material is posted here with permission of the IEEE. Such permission of the IEEE does not in any way imply IEEE endorsement of any of theUniversity of Pennsylvania's products or services. Internal or personal use of this material is permitted. However, permission to reprint/republish thismaterial for advertising or promotional purposes or for creating new collective works for resale or redistribution must be obtained from the IEEE bywriting to pubs-permissions@ieee.org. By choosing to view this document, you agree to all provisions of the copyright laws protecting it.

This paper is posted at ScholarlyCommons. http://repository.upenn.edu/hms/37For more information, please contact repository@pobox.upenn.edu.

Author(s)Norman I. Badler, John R. Clarke, Michael J. Hollick, Evangelos Kokkevis, Dimitris Metaxas, RamamaniBindiganavale, Bonnie L. Webber, Diane M. Chi, Nick Foster, Omolola Ogunyemi, and Jonathan Kaye

This conference paper is available at ScholarlyCommons: http://repository.upenn.edu/hms/37

MediSim:

SimulatedMedical Corpsmen and Casualties

for Medical Forces Planning and Training�

Norman I. Badler Bonnie L. Webber

John R. Clarke, M.D. Diane M. Chi

Michael J. Hollick Nick Foster

Evangelos Kokkevis Omolola Ogunyemi

Dimitri N. Metaxas Jonathan Kaye

Rama Bindiganavale

Center for Human Modeling and Simulation

Department of Computer and Information Science

University of Pennsylvania

Philadelphia, PA 19104-6389

Abstract

The MediSim system extends virtual environments(both local and network) to represent simulated medicalpersonnel interacting with simulated casualties. Ourtechnology fosters dual-use applications involving plan-ning, training, and evaluation of both medical corps-men and civilian EMTs. Behaviors and behavioralcontrol are being developed for the medical corpsmenthat will enable their actions on the digital battle�eldto conform to both military practice and medical proto-cols. From situationally-appropriate injury models, aset of physical and behavioral manifestations in a sim-ulated casualty will be determined and portrayed on athree-dimensional body.Keywords: Injury models, medical procedures,

computer simulation, virtual environments, physics-based modeling, medical decision support.

1 Introduction

The MediSim system extends virtual environmentsto represent simulated medical personnel interactingwith simulated casualties. Fig. 1 diagrams the majorcomponents of the MediSim system.

0Appeared in The National Forum: Military Telemedicine

On-Line Today. Research, Practice and Opportunities, IEEE

Computer Society Press, 1995.

� Existing human models are being extended toprovide close-range simulated casualties. Casual-ties must show medically signi�cant features suchas face, skeletal structures, and internal organs.Wound and injury types are being coupled prob-abilistically to the simulated battle�eld. Casu-alties must appear su�ciently real in terms ofboth physical behavior and appearance, in orderto motivate the trainee and his avatar to expedi-tiously but carefully provide medical intervention.The Jack R [3] human model is being extendedwith capabilities for appropriate injury display.A stealth human instructor can create simulatedinjuries and patient responses to treatment whilemonitoring the training exercise.

� The semi-automated medical corpsman (SAM)must be able to demonstrate rescue and medicalassessment procedures as a model for the traineeor to function as an independent agent duringa simulation. The SAM must demonstrate suf-�cient trauma care skills and practices in confor-mance with corpsman or EMT protocols to ren-der reasonable and responsible simulated medi-cal care. The Jack human model behavioral setis being augmented with a repertoire of relevantcorpsman behaviors such as taking vital signs tocon�rm and/or augment PSM data, immobiliz-ing suspected fractures, and stabilizing and trans-

Architecture of MediSIM System

Casualty Model

Jack Visualization

Corpsman Model(low level behaviors)

TraineeEvaluation

Semi-automatedbehavior(protocols)

Stealth Instructor

Trainee Participant

Wound/InjuryProbabilities

physical/behavioralmanifestations

informational/physical/behavioralresponses to corpsman actions

findings observationactions

(optional)

voice

I-port

feedback

DIS

Figure 1: Architecture of Proposed MediSim System

porting casualties with signs and symptoms ofshock. Task ordering will be determined by es-tablished medical corps or EMT protocols [5, 13].

� Virtual Environment technology will insure thefeeling of presence in the battle�eld environ-ment [14, 16] (Fig. 2). A VE embedding (avatar)of a corpsman must be produced, with physi-cal control over navigation and voice control overmedical procedures, allowing the corpsman to di-rect his/her avatar as if instructing a knowledge-able assistant.

� Evaluation measures will assess both quantitative(timing, amount of movement, change in patient'sphysiological state) and qualitative (care qual-ity, e�ectiveness) aspects of interactive corpsmantraining.

MediSim o�ers a number of signi�cant advances:

� Synthetic generation and visualization of woundsand injuries, along with their concomitant physi-cal, physiological, and behavioral manifestations.

� Model-based visualizations of injuries and vitalsigns that trigger corpsman decision-making andbehaviors implemented as a hierarchy of behaviornets.

� Task-based coordination of either multiple syn-thetic human �gures or combinations of synthetic�gures and VR avatars.

� Communication channel between the trainee andthe avatar modeled in an e�ective VE embedding.

Figure 2: Scene in the Battle�eld Environment

� The use of a \stealth" instructor to create, mon-itor, and e�ect injuries and dictate the out-comes of procedures or treatment response, placesthe MediSim project within the highly successfulparadigm of ight simulation but entails majorinnovation in the stealth's user interface.

� A dual-use but \military-ready" system is to bedesigned from the start with DIS compatibilitywhere possible. Both civilian EMT and mili-tary medical corpsmen procedures will be mod-eled within the same extensible framework.

In this paper, we �rst present an overview of theMediSim System and then discuss a prototype real-time implementation. Next, we discuss our model ofthe wounded soldier, including the generation of pene-trating injuries, casualty simulation, local wound sim-ulation, and physics-based modeling of medical pro-cedures. We conclude with a discussion of concurrentwork in medical decision support and developments inintegrated anatomical and physiological modeling.

2 The MediSim System

Our MediSim medical corpsman system consists ofseveral simulated individuals: one or more casualtieswith battle�eld or situation appropriate wounds, andhuman agents representing simulated medical corps-men who carry out procedures on the casualties.An agent may either be a semi-autonomous medic(SAM) or else an intentionally correlated embodiment(avatar) of a real participant's actions and commands.

In the latter case, the participant experiences the ca-sualty care situation with its concomitant dangersand challenges and may thus participate through theavatar as a second virtual agent observing, assisting,and commanding the SAM.

Casualties under corpsman care should display ap-propriate signs and manifestations of battle�eld in-jury, for purposes of corpsman training. Physiologi-cal state information is available by visual inspection,appropriate query, or correct performance of an as-sessment procedure. The medical corpsman behav-ioral repertoire needs to include protocol and proce-dures for rendering on-site emergency assessment andcare for a selection of such injuries. Consequently, thephysical and physiological e�ects and consequences ofthose procedures on the simulated casualty and hisinjuries are portrayed. During the course of the train-ing activity, the casualty's wounds and injuries as wellas his response to medical procedures can be proba-bilistically simulated { or controlled by a \behind thescenes" or \stealth" instructor analogous to the expe-rienced human trainer creating emergency situationsfor pilots during ight simulation.

The simulated medical corpsman will be able to actsemi-autonomously as a SAM, or through a live par-ticipant in an I-Port Virtual Environment interface asan avatar. The SAM will be able to navigate the dig-ital nanoterrain of the battle�eld, requesting and/orresponding to vital sign information transmitted bypersonnel status monitors (PSM). Protocols will di-rect the SAM to select the next casualty to attend to,using PSM, GPS, access route and threat data. Prox-imity to the casualty will enable physiological data tobecome available, triggering appropriate assessment,initial management, and evacuation procedures. Thephysical interactions between SAM and casualty willbe portrayed through real-time behaviors computedautomatically by the underlying human �gure simula-tion system.

For the live corpsman participant, the avatar willconform to input behaviors in carrying out navigation,and on verbal commandwill execute various behaviorsto carry out assessment, management, and evacuation.The participant is both observer and trainee, makingdecisions and giving commands to guide the physicallycapable, obedient, but passive avatar. Quantitativeevaluation measures will be examined to gauge thee�cacy of virtual training environments for medicalpersonnel.

2.1 Prototype Real-Time Implementa-tion

Early versions of portions of MediSim have beendemonstrated. In these con�gurations, pre-renderedmedical animation sequences were added to theJackML motion library. NPSNet (from the NavalPostgraduate School [14]) used the JackML to gener-ate human motions involving typical combat-orientedactivities, as well as specialized medical treatment an-imations. The participant in the exercises interactedwith the environment by using the Sarcos I-Port, aVR mobility simulator.

This system was tied into a standard DistributedInteractive Simulation (DIS) network. In this way,medic treatment scenarios could be performed in anenvironment populated by external DIS entities (al-though the medical animations themselves were notrendered on standard DIS image generators). Flagsneeded for medical animation were stored in unusedsegments of standard DIS Protocol Data Units; as thissystem evolves, new methods of transporting this datawill be developed.

2.2 Modeling the Wounded Soldier

Our basic approach to casualty modeling is to (1)generate a set of wounds and determine appropri-ate injuries on a simulated casualty either based onArmy probabilities (conditioned on weaponry, battle-�eld layout, etc.) or through penetration path mod-els (Section 2.2.1) ; (2) map them to likely �ndingsthrough information in our trauma management deci-sion support system, TraumAid (see Section 3.1); and(3) instantiate their physical and behavioral manifes-tations in a human �gure simulated in Jack and theirphysiological manifestations in simulated PSM data.

Given the weaponry in use, battle�eld layout, etc.,conditional probabilities specify the likelihood of dif-ferent wound(s). Given particular wounds, furtherconditional probabilities specify the types of injuries acasualty is likely to su�er. We will use these probabil-ities to generate simulated casualties. Alternatively,the stealth instructor can generate speci�c casualtytypes and responses (cf. Section 2.2.3).

Rules used in TraumAid to reason from �ndingsand test results forward to diagnostic and therapeu-tic goals can be inverted to map penetrating injuriesof the abdomen and chest, back to �ndings that inconnection with wound speci�cations give evidencefor those injuries. The problem that test results willnot be available can be solved in the same way as in

TraumAID's retrospective evaluation [8]: for diagnos-tic tests that were not performed during actual care, arealistic \default" database associated the likely resultof each test with each particular discharge diagnosis.

Findings can manifest through a casualty's physi-cal appearance or his behavior. Physical appearancecan be observed visually (e.g. distended abdomen,evisceration) or aurally (e.g., decreased breath sounds,stridor). Observation may be passive or require activ-ity on the part of the corpsman (e.g., tactile probingfor signs of tenderness and/or guarding). Behaviorsthat give evidence of injuries or degree of injury in-clude confusion, level of response to commands, gasp-ing muscle spasms, etc.

Although the current TraumAid system only coverspenetrating injuries to the chest and abdomen, smallTraumAid rule sets for upper and lower extremity in-juries developed for the Navy's Submarine Center atGroton [7] will allow us to generate at least a subsetof appropriate �ndings for such injuries as well.

By the end of the current year, we expect todemonstrate casualties exhibiting physical manifesta-tions of selected integument and musculo-skeletal in-juries. During the second year, casualties will exhibitboth physical and behavioral manifestations of addi-tional integument, musculo-skeletal and internal in-juries; for example, swelling, pale skin color, exposedviscera, breathing anomalies, muscle spasms, or facialexpressions.

2.2.1 Modeling Penetrating Injuries

Determining the anatomical and physiological impactof penetrating wounds involves relating knowledge ofanatomy with possible projectile penetration paths.Our penetration path module assists in evaluatingpenetrating injuries to the chest and abdomen, andcomplements MediSim's injury model. Delp [10] ad-dresses penetrating injuries to the legs.

Penetrating injuries caused by gunshot or stabwounds are modeled using a rotatable 3D graphicalmodel of the torso with the appropriate anatomicalstructures. External wounds corresponding to eithergunshot or stab wounds can be entered onto any loca-tion on the surface of the 3D body model. In the caseof ballistic injuries, the presence of multiple entry andexit wounds complicates the process of determiningwhich organs have been injured.

A combinatorial analysis of the surface woundsleads to various penetration path hypotheses. For anygiven pair of wounds, we de�ne a wound path space asthe space of possible paths a bullet may have takenfrom one wound to the other. Similarly, for a sin-

gle external wound, a wound path space describes thespace of possible trajectories from the wound to a bul-let lodged in the body. A penetration path hypothesismay consist of one or more wound path spaces, de-pending on the number of wounds.

Once all the possible pairings of wounds for a givenset of penetrating injuries have been determined, thesystem establishes which anatomical structures are af-fected for each wound path space. By displaying eachpairing of penetration path and injured organs for agiven set of wounds, the system provides a visual cueto their consequences. This can help in bridging thegap between knowledge of the anatomy involved in aparticular injury and the physiological manifestationsassociated with that injury.

2.2.2 Casualty Simulation

Casualty simulation begins with the speci�cation ofinjuries sustained by the victim. Currently, the mod-eled injuries are (1) chest injury resulting in a ten-sion pneumothorax and (2) abdominal injury result-ing in abdominal bleeding, hemothorax, and pericar-dial tamponade. The simulation displays plots of thevictim's vital signs over time. The plotted values ofthe vital signs use the rankings given in the TraumaScale [6] endorsed by the American Trauma Societyfor estimating the injury severity. The trauma scalere ects basic assessments of the respiratory, circula-tory, and central nervous systems. Fig. 3 shows theTrauma Scale rating system.

A Jack human �gure portrays the physical andbehavioral manifestations of its injuries and re-sponds appropriately to commands representing var-ious diagnostic and therapeutic procedures. Thefollowing diagnostic queries result in a text re-sponse written to the screen: respiratory-rate?,blood-pressure?, capillary-refill?, name?, id?.Visual responses from the Jack �gure result fromassessing respiratory expansion, eye opening abil-ity, and motor response. These commands in-clude: open-your-eyes, apply-sternal-pressure,move-your-foot, and squeeze-my-hand. The usermay also use the following commands for administer-ing treatments: needle-aspiration, give-fluids,and occlusive-dressing. The commands for assess-ing the situation are: are-you-ok?, what-happened?,do-you-have-pain?, andwhere-does-it-hurt?. The commands and queriesare currently not implemented as procedures carriedout by the (semi-)autonomous Jack medic, but rathertext input commands to the interface which invokean appropriate patient response. The simulation will

Respiratory 10-24/min 4Rate 24-35/min 3

36/min or greater 21-9/min 1None 0

Respiratory Normal 1Expansion Retractive 0Systolic 90 mmHg or greater 4Blood 70-89 mmHg 3Pressure 50-69 mmHg 2

0-49 mmHg 1No Pulse 0

Capillary Normal 2Re�ll Delayed 1

None 0

Glasgow Coma Scale

Eye Spontaneous 4Opening To Voice 3

To Pain 2None 1

Verbal Oriented 5Response Confused 4

Inappropriate Words 3Incomprehensive Words 2None 1

Motor Obeys Commands 6Response Localized Pain 5

Withdraw Pain 4Flexion (pain) 3Extension (pain) 2None 1

Figure 3: Trauma Scale

be extended to display additional physical and be-havioral manifestations; for example, distended neckveins, cyanosis, thrashing, and other general expres-sions of pain.

The model is built using Parallel Transition Net-works (PaT-Nets), a package for creating and runningparallel state-machines [2, 4]. There are two maintypes of networks that compose the casualty model:

� Vital sign networks which generate the visual ef-fects of changes in trauma scale measurementsand respond to user input, and

� Injury networks that specify the physiologicalchanges resulting from speci�c medical condi-tions.

Each modeled vital sign has a network containingmethods to produce appropriate visual and textual re-sponses to user commands re ecting the patient's cur-rent state (trauma score values). A controller networkensures that the values remain in their valid ranges.

Physiological conditions that result from injury aremodeled as individual PaT-Nets. These networksspecify the transitions in the vital sign controller net-work for the trauma scale measurements. The pa-tient's condition, re ected in the trauma score, is setto deteriorate or improve at the appropriate times,depending on the performed therapeutic intervention(or lack thereof). Currently, the modeled conditionsinclude tension pneumothorax, hemothorax, pericar-dial tamponade, and abdominal bleeding. A casualtymay receive multiple conditions as result of injury. Forinstance, a penetrating abdominal wound may resultin hemothorax, pericardial tamponade, and abdom-inal bleeding. When multiple conditions occur, thelowest value of each trauma scale value is used as theoverall value. The current networks were built basedon estimates from one of the authors, an experiencedtrauma surgeon (Clarke).

2.2.3 Case Study: Tension Pneumothorax

Fig. 4 shows the tension pneumothorax and needleaspiration PaT-Nets. The machine starts in the BaseNode. A clock starts at the time of injury. After spe-ci�c time intervals (1, 3, 5, 8, 12, and 15 minutes),the PaT-Net transitions to a new state where the ap-propriate vital signs are decremented. Thereafter, thePaT-Net immediately returns to the Base Node. Ifthe patient is given a needle aspiration, the tensionpneumothorax PaT-Net exits and spawns a needleaspiration PaT-Net. The needle aspiration PaT-Netstarts in its Base Node. At one and two minute in-tervals, it transitions to a new state and incrementsthe appropriate vital sign values. Fig. 5 shows theplot of the vital signs re ecting the Central NervousSystem when the patient develops a tension pneumo-thorax and receives no treatment. Fig. 6 shows theCentral Nervous System plot when the patient receivesa needle aspiration 9 minutes after the time of injury.

2.3 Localized wound simulation

Wound simulation is considered important in thetraining of �rst response emergency teams for two rea-sons;

1. The physical appearance and behavior over timeof the wound site are vital clues for diagnosis.

BaseNode(tp)

RR<−3

RR<−2RE<−0

BP<−3CR<−1EO<−3VR<−4BP<−2

EO<−2VR<−3MR<−5

RR<−1BP<−1CR<−0EO<−1VR<−2MR<−4

RR<−0BP<−0VR<−1MR<−1

spawn needle

aspirationpatnet

and exit

BaseNode(na)

inc−BPinc−CRinc−EOinc−VRinc−MR

inc−RRinc−REinc−BPinc−CRinc−EOinc−VRinc−MR

t=1

t=3

t=5

t=8t=12

t=15

give−nadefault

default

defaultdefault

default

default

defaultdefault

1 minelapsed

2 minelapsed

Figure 4: Tension Pneumothorax and Needle Aspira-tion PaT-Nets

2. The size, shape, and location of wounds in uencethe rates at which physiological changes occur inthe casualty.

The goal of this part of the project is to combine anaccurate model of the wounding mechanism, i.e. thedamage caused by a projectile, with a visual portrayalof the victim's localized physical response. Bleedingis one of the most important aspects of that responsebecause blood loss not only impacts the extent of theinjury but may also cause secondary trauma such asparalysis. A system for modeling and animating liquidphenomena [12] is being developed to simulate inter-nal and external bleeding due to penetrative wounds.The method is based around the Navier-Stokes equa-tions which can be used to describe the motion of aviscous uid. This model has been extended to give re-alistic behavior to blood including clotting and waterbridging. Where possible, computationally expensiveparts of the uid calculation have been simpli�ed togive the high frame rate necessary for the real-timerequirement of the MediSim application.

2.4 Physics-based Modeling of MedicalProcedures

Physical realism plays an important role in thesimulation of the medical corpsmen procedures. The

Figure 5: Tension Pneumothorax Plot { No Treatment

Figure 6: Tension Pneumothorax Plot { Needle Aspi-ration at 9 minutes

quality of the motion animation can be enhanced byadding dynamical properties to the humans and ob-jects present in the scenes. After assigning realisticmass and inertia to both the body parts of the injuredhumans and the medical instruments and tools, natu-ral looking motion can be generated though a dynamicsimulator. The corpsman can interact with the envi-ronment and perform the speci�ed operations by ap-plying (mainly through his hands) external forces di-rectly to the injured person. The e�ect of these forceswill automatically be computed by the dynamic simu-lator and the resulting motion of the parts and objectsinvolved will be both physically correct and convinc-ing. The advantages o�ered from dynamic simulationare not limited only to the visual realism of the ani-mation. Given accurate strength limits for the humanbody muscles and joints, the feasibility of certain op-erations that the corpsman performs can be judged. Ifmore force to achieve a certain result (such as movingthe injured body) is needed than the corpsman canapply then a request for extra help can be issued. Afast recursive dynamic simulator [11] specially suitedfor the e�cient animation of articulated �gures is ap-plied to compute the motions in real time. The in-ertial properties of objects involved in the scenes caneither be automatically inferred by assuming a presetmass density for each object or can be set o�-line bya user. Dynamic simulation adds to the realism ofsimulated procedures, provides quantitative measuresfor the feasibility of operations on the injured humans,and automates the process of creating new simulationscenarios.

Patient response will conform dynamically andprobabilistically to a set of implemented injury types.Using Metaxas' techniques of physics-based modelingand dynamic deformations [15], we will be able to dosome dynamic computation of the motion of unsup-ported body parts and restraint of distended soft or-gans/tissues.

The casualty model generates simulations of vari-ous battle�eld injuries. The model produces appropri-ate physical, physiological, and behavioral responsesto both injuries and medical intervention.

3 Related Activities

3.1 Medical Decision Support and Treat-ment Planning

We have developed and retrospectively validated adecision-support system for the initial de�nitive man-agement of multiple trauma (penetrating injuries of

the chest and abdomen) in Emergency Centers [8, 18].Deployment in the Emergency Center at the MedicalCollege of Pennsylvania for the purpose of prospectiveevaluation began in April 1995. An earlier version ofthe TraumAid system was adapted to use by medicalcorpsmen on submarines and extended to cover basiccare of injuries to the extremities as well [7].

TraumAID was designed to complement the cov-erage of ATLS procedures [1] and conform to the in-formation needs of local, state and national traumaregistries. We are now following developments in therecently established Trauma Care Information Man-agement Systems (TCIMS) TRP consortium, whosegoal is to apply telecommunications, information anddecision support technologies to improving the speedand e�ectiveness of pre-hospital trauma care. We seethat through interfacing with TCIMS, information col-lected pre-hospitally can be used to reduce Traum-AID's in-hospital data collection needs and position itto provide decision support for de�nitive managementas soon as the patient enters the Emergency Center.

3.2 Integrated Anatomical and Physio-logical Modeling

The TrauMAP project aims to link physiologicalmodels that express mechanical behavior (and ulti-mately chemical and neurological functions) with avisually- and structurally-accurate anatomical model.Work to date has focussed on modeling respiratorymechanics, in order to demonstrate both quiet, normalbreathing and pathologies in the form of pneumotho-races. If one were only interested in such demonstra-tions per se, scripted animation might su�ce. How-ever, since the aim is to simulate and ultimately pre-dict interactions among body systems, physical ob-ject models are needed that respond to and transmitforces. Because the anatomical components are prin-cipally soft tissue, that response must include objectdeformation. Thus, the work to date has involved in-tegrating a 3D deformable simulation of the lungs andthoracic space with physiological process equations,expressed as ordinary di�erential equations. Thiswork is described in more detail in [9].

Acknowledgments

This work is being undertaken in conjunction withSharon Stans�eld of Sandia National Laboratoriesand Michael Zyda and David Pratt of the NavalPostgraduate School. This research is partially sup-ported by ARPA Biomed Program DAMD17-94-J-4486; NLM N01-LM-4-3515; DMSO DAAH04-94-

G-0402; U.S. Air Force DEPTH through HughesMissile Systems F33615-91-C-0001; ARO DURIPDAAH04-95-1-0023; Army AASERT DAAH04-94-G-0220; ARPA AASERT DAAH04-94-G-0362; and NSFCISE CDA88-22719.

References

[1] American College of Surgeons, Advanced TraumaLife Support, Chicago: American College of Sur-geons, 1984.

[2] Badler, N. and Becket, W., \Integrated Behav-ioral Agent Architecture," Proc. 3rd Conf. onComputer Generated Forces and Behavior Rep-resentation, Orlando, FL, pp. 57-68, March 1993.

[3] Badler, N., Phillips, C. and Webber, B., Simu-lating Humans: Computer Graphics, Animation,and Control, Oxford: Oxford University Press,1993.

[4] Badler, N., Webber, B., Becket, W., Geib, C.,Moore, M., Pelachaud, C., Reich, B., and Stone,M., \Planning for Animation," in D. Thalmannand N. Magnanat-Thalmann (eds.), ComputerAnimation, New York: Prentice Hall Inc., 1995.

[5] Brady, Emergency Care, 5th edition.

[6] Champion, H.R., Sacco, W.J., Carnazzo, A.J.,et. al., \Trauma Score," Critical Care Med. 9(9),pp. 672{676, 1981.

[7] Clarke, J.R., Niv, M., Webber, B.L., Fisherkeller,K., Southerland, D.G. and Ryack, B., \Traum-Aid: A Decision Aid for Managing Trauma atVarious Levels of Resources," Proc. 13th AnnualSymp. on Computer Applications in Medical Care(SCAMC), Washington D.C., pp. 1005-1006 (ab-stract), November 1989.

[8] Clarke, J.R., Rymon, R., Webber, B., Hayward,C., Santora, T., Wagner, D. and Ru�n, A., \TheImportance of Planning in the Provision of Medi-cal Care," Medical Decision Making 13(4), p. 383(abstract), October-December 1993.

[9] DeCarlo, D., Kaye, J., Metaxas, D., Clarke,J.R., Webber, B. and Badler, N., \Integrat-ing Anatomy and Physiology for Behavior Mod-eling," in Medicine Meets Virtual Reality III:Interactive Technology and the New Paradigmfor Healthcare, R.M. Satava, K. Morgan, H.B.

Sieburg, et al eds., Amsterdam: IOS Press,pp. 81-87, 1995.

[10] Delp, S., \Battle�eld Casualty Care Using theVirtual Human," this volume.

[11] Featherstone, R., \The Calculation of RobotDynamics Using Articulated-Body Inertias," In-ternational Journal of Robotics Research, 2(1),Spring 1983.

[12] Foster, N. and Metaxas, D., \Realistic Animationof Liquids," Submitted to GMIP, 1995.

[13] Henry, M. and Stapleton, E., EMT PrehospitalCare, Philadelphia: W.B. Saunders Co., 1992.

[14] Macedonia, M.R., Zyda, M.J., Pratt, D.R.,Barham, P.T. and Zeswitz, P., \NPSNET: A Net-work Software Architecture for Large-Scale Vir-tual Environments," Presence: Teleoperators andVirtual Environments 3(4), pp. 265-287, 1994.

[15] Metaxas, D. and Koh, E., \E�cient Shape Repre-sentation Using Deformable Models with LocallyAdaptive Finite Elements," Geometric Methodsin Computer Vision II, B.C. Vemuri (ed.), Proc.SPIE 2031, pp. 160-171, San Diego, CA, July1993.

[16] Stans�eld, S.A., \Distributed Virtual RealitySimulation System for Situational Training,"Presence: Teleoperators and Virtual Environ-ments 3(4), pp. 360-366, 1994.

[17] U. S. Army Medical Department (AMEDD)Videotapes, Perform Patient Assessment, FortSam Houston, TX, November 1993.

[18] Webber, B., Rymon, R. and Clarke, J.R., \Flex-ible Support for Trauma Management throughGoal-directed Reasoning and Planning," Arti�-cial Intelligence in Medicine 4(2), pp. 145-163,April 1992.