400 Commonwealth Drive, Warrendale, PA 15096-0001 U.S.A. Tel: (724) 776-4841 Fax: (724) 776-5760 Web: www.sae.org

SAE TECHNICALPAPER SERIES 2006-01-0961

Optimizing Vehicle Occupant Packaging

Matthew B. ParkinsonEngineering Design, The Pennsylvania State University

Matthew P. ReedUniversity of Michigan Transportation Research Institute

Reprinted From: Reliability and Robust Design in Automotive Engineering 2006(SP-2032)

2006 SAE World CongressDetroit, Michigan

April 3-6, 2006

The Engineering Meetings Board has approved this paper for publication. It has successfully completed SAE's peer review process under the supervision of the session organizer. This process requires a minimum of three (3) reviews by industry experts.

All rights reserved. No part of this publication may be reproduced, stored in a retrieval system, or transmitted, in any form or by any means, electronic, mechanical, photocopying, recording, or otherwise, without the prior written permission of SAE.

For permission and licensing requests contact:

SAE Permissions400 Commonwealth DriveWarrendale, PA 15096-0001-USAEmail: permissions@sae.orgTel: 724-772-4028Fax: 724-776-3036

For multiple print copies contact:

SAE Customer ServiceTel: 877-606-7323 (inside USA and Canada)Tel: 724-776-4970 (outside USA)Fax: 724-776-0790Email: CustomerService@sae.org

ISSN 0148-7191Copyright © 2006 SAE InternationalPositions and opinions advanced in this paper are those of the author(s) and not necessarily those of SAE. The author is solely responsible for the content of the paper. A process is available by which discussions will be printed with the paper if it is published in SAE Transactions.

Persons wishing to submit papers to be considered for presentation or publication by SAE should send the manuscript or a 300 word abstract to Secretary, Engineering Meetings Board, SAE.

Printed in USA

ABSTRACT

Occupant packaging practice relies on statistical modelscodified in SAE practices, such as the SAE J941 eyel-lipse, and virtual human figure models representing indi-vidual occupants. The current packaging approach pro-vides good solutions when the problem is relatively un-constrained, but achieving good results when many con-straints are active, such as restricted headroom and sight-lines, requires a more rigorous approach. Modeling driverneeds using continuous models that retain the residualvariance associated with performance and preference al-lows use of optimization methodologies developed for ro-bust design. Together, these models and methods facil-itate the consideration of multiple factors simultaneouslyand tradeoff studies can be performed. A case study in-volving the layout of the interior of a passenger car ispresented, focusing on simultaneous placement of theseat and steering wheel adjustment ranges. Tradeoffs be-tween adjustability, driver accommodation, and exterior vi-sion are explored under this paradigm. These results arecontrasted with those obtained using boundary manikins.

INTRODUCTION

Vehicle occupant packaging is the process of laying outthe interior of a vehicle to achieve the desired levels of ac-commodation, comfort, and safety for the occupants (Roe1993). The primary focus in occupant packaging is thedriver’s workstation. The driver package usually refersto the locations and adjustment ranges of the steeringwheel and seat with respect to the pedals, but also en-compasses the physical locations of controls and displayswith which the driver interacts. Analysis of interior andexterior driver vision zones, both direct and indirect (us-ing mirrors), is also typically considered part of packagingpractice. Packaging is begun early in the vehicle designprocess, although often after the body exterior contour

and window openings have been designed, as the firststep in the design of the interior. The package is iteratedupon through the design process as constraints change.

The objectives of the packaging process are usuallystated in terms of percentage accommodation on par-ticular measures. Accommodation is quantified as thefraction of the driver population achieving some targetedlevel of fit or comfort (Roe 1993). For example, the seattrack adjustment range is often selected so that 95% ofdrivers are able to sit with the seat position that theyprefer. Another common goal is to ensure that 95% ofdrivers achieve a minimum upward vision (upvision) an-gle through the windshield.

Beginning in the late 1950s, the Society of Automo-tive Engineers began considering standardized tools andprocedures for packaging. SAE Recommended Prac-tice J826, first approved in 1962, defined a weightedthree-dimensional manikin for measuring seats and a two-dimensional template with a similar profile for use onpackage drawings (SAE 2005). The manikin, known asthe H-point machine, defines and measures the locationof the H-point, a reference point that approximates the hiplocation of a person sitting in the seat (see Reed et al.1999). The template creates a standardized, schematicvisualization of a driver with long legs, originally for pur-poses of ensuring adequate legroom for tall men (Geof-frey 1961). A new H-point manikin was developed in the1990s (Reed et al. 1999) and is now the basis for SAEJ4002 (SAE 2005).

In addition to the measurement tools, the SAE Recom-mended Practices relating to packaging are focused onpercentile accommodation models created by statisticalanalysis of data on driver behavior. SAE J941 introducedthe first percentile accommodation model, the eyellipse,in the early 1960s (Meldrum 1965). The eyellipse is a

1

2006-01-0961

Optimizing Vehicle Occupant Packaging

Matthew B. Parkinson Engineering Design, The Pennsylvania State University

Matthew P. Reed University of Michigan Transportation Research Institute

Copyright © 2006 SAE International

graphical construction that describes the expected distri-bution of driver eye locations, approximated as a three-dimensional normal distribution. The eyellipse, which hasbeen used extensively for performing vision analyses overthe last four decades, was recently upgraded to repre-sent contemporary driver populations and to take into ac-count the effects of steering wheel position on eye location(Manary et al. 1998). Other important statistical modelsin SAE Recommended Practices include the seating ac-commodation model in SAE J1517 (now superceded bySAE J4004 for passenger cars and light trucks); the driverreach curves in SAE J287; and the driver head clear-ance contour in SAE J1052. In each case, the model pro-vides a geometric design guide that represents a specifiedpercentage of the relevant measure from a population ofdrivers.

An increasingly common approach to occupant packag-ing employs human figure models, rather than percentileaccommodation models, to represent driver requirements.The use of kinematic linkage models of the human form(manikins) to represent the size, shape, and posture of ve-hicle occupants in design dates at least to the 1950s (cf.Dempster 1955). The use of three-dimensional computergraphics models of humans for vehicle interior design andassessment, as with many other engineering software ap-plications, has followed the development of low-cost com-puters and the transition from paper to 3-D computer mod-eling for vehicle design. Early human modeling softwareprograms such as SAMMIE (Porter et al. 1993) have beenjoined in the marketplace by Ramsis, Jack, and Safework,among others (Chaffin 2001). These digital human mod-els (DHM) are now widely used for vehicle interior de-sign and have replaced the SAE packaging tools in somecompanies, particularly in the commercial-vehicle indus-try (Loczi 2000).

Roe (1993) presents a useful overview of the rel-ative strengths and weaknesses of the percentile-accommodation and manikin-based approaches. Themodels provide quantifiable accuracy and precision for theapplicable variables (e.g., eye location in normal drivingposture) and are fundamentally population models. Thatis, they describe percentiles of a population, not the be-havior of any individual within the population. The primarydisadvantage of the percentile accommodation models istheir high level of abstraction. For example, communi-cating information about the need to improve visibility forpeople with low eye positions can be facilitated using amodel that looks like a small woman, even if the eyellipseis needed for an accurate quantitative vision analysis.

An important limitation of the percentile accommodationmodels (e.g., the eyellipse in J941 and the seating ac-commodation model in J4004) is that they are essentiallyunivariate, dealing with a single measure of interest (seatposition, eye location). In addition, only a subset of thepotential geometric factors influencing driver posture areincluded in the models. For example, the potential influ-

ence of a low roof on driver eye location, through restric-tive headroom, is not taken into account in J941.

Figure models provide good face validity and also pro-vide the capability of responding to kinematic constraintsthat are not represented in the population accommodationmodels. The effects of restricted headroom could be rep-resented by requiring the figure models to maintain a gapbetween their heads and the roof. A large number of fig-ure models would be needed to attain good estimates ofpopulation accommodation. In an attempt to reduce thenumber of figure model analyses that must be performed,analysts frequently select a small number of boundarymanikins that span a large percentage of the range ofbody dimensions in the target population. Methods forselecting boundary manikins range from simplistic ”5th-percentile female and 95th-percentile male” approachesto sophisticated techniques using factor analysis (Bittner2000).

However, the boundary manikin approach does not gen-erally provide accurate assessments of accommodation.As typically used in software packages such as Jack andRamsis, only the variance in posture attributable to bodydimensions and vehicle geometry is included. Reed andFlannagan (2001) showed that variance in outcomes ofinterest (e.g., eye location and seat position) that is notassociated with body dimensions and vehicle geometrymust be included to obtain accurate assessments of pop-ulation accommodation from figure model analyses. How-ever, this requires a simulation method that allows thisresidual variance to be taken into account. Recently,Parkinson et al. (2005) introduced a new approach todriver packaging based on an advanced use of the figure-model approach in the context of an optimization frame-work. Using the example of truck cab packaging, driverswere sampled randomly from an anthropometric distribu-tion. At each design evaluation, the drivers were posturedin the vehicle using statistical posture prediction mod-els. A random component was added to each predicteddegree-of-freedom based on the findings from the driver-posture studies used to develop the posture-predictionmodels. The optimization framework was used to deter-mine the optimal seat track locations under the influenceof geometric constraints, including a short cab length andlow roof height.

The present paper extends these methods to SAE Class-A vehicles (passenger cars and light trucks with designseat heights less than 405 mm) and compares the effi-cacy of the method to the boundary manikin approach forseveral vehicle design problems.

METHODS

PARAMETERIZING THE DRIVER PACKAGE Figure 1shows the reference points and variables used to describethe driver package. All analyses were performed in two di-mensions (side view). The steering wheel was described

2

by a pivot location, range of telescope, and range of an-gle. The seat track adjustment range (H-point travel path)was defined by the location of the full-forward, full-downposition relative to the accelerator heel point (AHP) andthe horizontal and vertical adjustment ranges. The trackwas fixed at 6 degrees downward from horizontal, a typi-cal value for passenger cars. The height of the floor androof were defined with respect to the ground, and a roofthickness was assigned. Upward and downward visionrestrictions were represented by three points: the upperdaylight opening (DLO), the cowl point at the base of thewindshield, and the hood point.

Among the dependent measures of interest were two vi-sion angles calculated as shown in Figure 1. Upvision an-gle was calculated as the angle with respect to horizontalof the vector from the drivers eye location passing throughthe upper daylight opening point. Downvision angle wascalculated using the vector through either the cowl pointor hood point, whichever was most restrictive. Many otherdesign variables and accommodation measures are im-portant for packaging, but the set in Figure 1 representsthe core of the driver packaging issues.

DRIVER POPULATION For the current analysis, thedriver population was based on an analysis of data from a1988 anthropometric survey of U.S. Army personnel (Gor-don et al. 1989). The ANSUR data are widely used for an-thropometric analyses because the data are publicly avail-able and there are a large number measures in the dataset. Table 1 lists the body dimensions of interest for thecurrent study along with summary statistics. Drivers wererepresented by gender, stature (erect standing height),erect sitting height, and body mass index (BMI). BMI iscalculated as the body mass in kilograms divided by thestature in meters, squared. The result is a measure ofweight-for-stature that is less correlated with stature thanis body weight.

A population of 500 men and 500 women was generatedby sampling randomly from a multivariate normal distri-bution for each gender characterized by the means andstandard deviations of stature, sitting height, and the natu-ral log of BMI. BMI was transformed to achieve better rep-resentation by the normal-distribution approximation. Toreduce correlation among the variables, the ratio of erectsitting height to stature was used in place of erect sittingheight. Table 2 lists the covariance matrices for the maleand female distributions.

Table 1: Means (standard deviations) of AnthropometricVariables for the Target Population.

Males Females

Stature (mm) 1755 (66.8) 1629 (63.6)Erect Sitting Height (mm) 914 (35.6) 852 (34.9)

Erect Sitting Height / Stature 0.521 (0.0144) 0.523 (0.0147)log(BMI) 3.23 (0.118) 3.14 (0.113)

BOUNDARY MANIKINS For comparison with therandomly-sampled population, a set of 14 male and 14female boundary manikins was generated. A typicalapproach to using boundary manikins in anthropomet-ric analyses follows these steps (HFES 300 Committee2004):

1. identify anthropometric dimensions that are related tothe accommodation measures of interest;

2. represent the anthropometric variance in these di-mensions using a multivariate normal distribution;

3. construct an ellipsoid in this multi-dimensional spacethat encloses the same percentage of the anthropo-metric space as the target accommodation level forthe design (e.g., 95 percent); and

4. sample individual cases (boundary manikins) fromthe surface of this ellipsoid, usually a minimum oftwo per anthropometric dimension (six or more for athree-dimensional anthropometric space).

The underlying assumption of the boundary manikinmethod is that accommodating individuals whose body di-mensions collectively span a particular percentage of theanthropometric space will accommodate at least that per-centage of actual users. Because the number of bodydimensions of interest is often higher than three, princi-pal component analysis (PCA) is often used to reducethe dimensionality, often to three components (cf. Bit-tner 2000). For the vehicle driver accommodation prob-lem, people might choose leg length, buttock-popliteallength, functional arm reach, sitting height, seated eyeheight, and abdomen depth, among others. Manary et al.(1999) showed that most of the variance in anthropomet-ric measures potentially related to driver accommodationwere accounted for by just three variables: stature, sit-ting height, and body weight. Seidl (1994) identified threesimilar measures as accounting for most of the variancein anthropometric measures relevant to driver accommo-dation, substituting waist circumference for body weight.

Table 2: Covariance Matrices for Multinormal Approxima-tions to Male and Female Anthropometric Distributions.

Males

Stature SH/S* log(BMI)Stature 4463 -0.3190 0.1008SH/S -0.3190 1266 0.00011

log(BMI) 0.1008 0.00011 0.0138

Females

Stature SH/S* log(BMI)Stature 4045 -0.2692 -0.4531SH/S -0.2692 0.00022 0.00010

log(BMI) -0.4531 0.00010 0.0127*ratio of erect sitting height to stature

3

Figure 1: Dimensions and reference points used in package optimization.

For the current analysis, the knowledge of the postur-ing models allows a more accurate selection of boundarymanikins than would ordinarily be possible. Specifically,the posture models depend only on stature, erect sittingheight, and body mass index. Consequently, only thesethree measures were included in the selection of bound-ary cases. As noted above, two transformations were per-formed to simplify the process by reducing correlationsamong the variables. First, the ratio of erect sitting heightto stature was substituted for erect sitting height. The logtransform of BMI was also used to reduce correlation withstature and to achieve a better representation by the nor-mal approximation.

For each gender, boundary manikins were selected tolie on the surface of an ellipsoid that enclosed 95 per-cent of the population, approximated by a multivariate nor-mal distribution with the covariance matrices given in Ta-ble 2. If accommodation is determined on the boundariesof the anthropometric distribution, which is the case forthe current analysis, then selecting cases in this manneris appropriate (HFES 300 Committee 2004). Boundarymanikins were chosen at the extremes on each axis ofthe enclosing ellipse (six manikins) and at eight interme-diate points between axes, as shown in Figure 2. Usingthis approach, fourteen manikins were generated for eachgender. Table 3 lists their body dimensions.

For application of these boundary manikins to human fig-ure models, calculation of all of the body segment dimen-sions would be required. These calculations are oftenperformed using multivariate regression or through factoranalysis (Bittner 2000). The current analysis required onlyone additional variable to be predicted, namely the heightof the head above the eyes. This value was computed for

each boundary manikin using a regression on erect sittingheight in the data from ANSUR.

Table 3: Body Dimensions of the 28 Boundary Manikins.

Gender Stature SH* SH/S** log(BMI) Mass BMI(mm) (mm) (kg) (kg/m2)

F 1766 866 0.490 3.049 65.8 21.1F 1659 855 0.515 3.444 86.3 31.3F 1738 951 0.547 3.147 70.3 23.3F 1492 829 0.556 3.239 56.8 25.5F 1599 849 0.531 2.844 43.9 17.2F 1520 758 0.499 3.141 53.4 23.1F 1469 782 0.533 3.024 44.4 20.6F 1595 894 0.560 3.027 52.5 20.6F 1505 787 0.523 3.370 65.8 29.1F 1630 898 0.551 3.374 77.6 29.2F 1628 806 0.495 2.914 48.8 18.4F 1753 917 0.523 2.918 56.9 18.5F 1663 808 0.486 3.261 72.1 26.1F 1789 918 0.513 3.264 83.7 26.2M 1904 928 0.487 3.187 87.8 24.2M 1791 935 0.522 3.552 111.9 34.9M 1649 821 0.498 3.273 71.7 26.4M 1606 890 0.554 3.271 67.9 26.3M 1719 893 0.519 2.906 54.0 18.3M 1861 1012 0.544 3.185 83.8 24.2M 1709 944 0.552 3.042 61.2 20.9M 1587 834 0.526 3.092 55.4 22.0M 1751 970 0.554 3.414 93.2 30.4M 1628 859 0.528 3.465 84.8 32.0M 1882 967 0.514 2.993 70.6 20.0M 1759 857 0.487 3.044 64.9 21.0M 1923 991 0.515 3.366 107.2 29.0M 1801 881 0.489 3.416 98.7 30.5

*erect sitting height**ratio of erect sitting height to stature

4

Figure 2: Schematic depiction of boundary manikin definition for men in three orthogonal views. The ellipsoid encloses 95%of the multivariate normal distribution for the three defining variables. Sampled boundary manikins (a total of 14) are shownas dots on the ellipsoid.

EXPERIMENTAL POPULATIONS Design analyseswere conducted using three populations of simulateddrivers. The population of 1000 drivers sampled randomlyfrom the anthropometric distributions for each gender,and including random variance in posture-selectionbehavior, was termed RS (random sample). The groupof 28 boundary manikins (with no associated randomvariance), was termed BM. To help to clarify the role ofrandom postural variance in determining accommodation,a third population was defined by instantiating each of the28 boundary manikins 10 times, each time with randomlysampled postural variance values. This population wastermed BMR (boundary manikin—random).

POSTURING FUNCTIONS Each sampled driver waspostured in the vehicle using the Cascade PredictionModel (CPM) described in Reed et al. (2002). The CPMis a cascaded series of regression models based on a sta-tistical analysis of driver posture data in a wide range ofvehicle configurations. The posturing process proceededas follows: Based on the specified steering wheel pivotlocation, the mean expected preferred location for a driverwith the specified anthropometry was calculated. The de-sired steering wheel position was censored (i.e., limited)to the available range of tilt and telescope. The resultingsteering wheel location and body dimensions were usedalong with body dimensions to calculate the desired seatposition (fore-aft and vertical). Seat position was also cen-sored to the available range. Hip location was calculatedas an offset from seat H-point, and eye location was cal-culated at a distance and angle relative to hip location.The prediction of each degree of freedom included an ap-propriate random variance component sampled from anindependent normal distribution for each driver (Reed etal. 2002).

DEFAULT VEHICLE DESIGN The design scenarioswere examined using a generic vehicle package as astarting point. Table 4 lists the parameter values. Referto Figure 1 for definitions.

PACKAGE OPTIMIZATION APPROACH Figure 3shows the general optimization method schematically.There are four primary components: vehicle definition,population definition, posturing model, and optimization.

For the current analysis, the horizontal and vertical loca-tion of the seat track (x and z) as well as its length andvertical adjustability (∆x and ∆z) are the design variables,x. Depending on the scenario, one of three driver popula-tions is selected (BM, BMR, or RS). A preferred posture isdetermined for each of the manikins in the selected pop-ulation, yielding a preferred seat location (xpref , zpref ). Ifthe seat can be adjusted to the desired location, the driveris said to have been accommodated. When the desiredlocation is not within the available range of seat adjust-ment, the nearest feasible location, (xact, zact), is calcu-lated and the spatial difference between the two locationsis retained as a measure of disaccommodation. The dis-accommodation for a single driver is then

d =√

(xpref − xact)2 + (zpref − zact)2. (1)

The total number of accommodated drivers, NA, for theselected population is obtained by summing the numberof drivers for whom d = 0. The percentage of accommo-dated drivers, FA is given by the equation

FA = 100NA

n, (2)

Table 4: Initial vehicle dimensions.Dimension* x (mm) z (mm)

Accelerator Heel Point (AHP) 0 400Steering Wheel Pivot re AHP 325 570Telescope Range from Pivot 175 225Steering Wheel Angle Range (deg) 10 40Upper Daylight Opening 500 1400Steering Wheel Diameter 355 -Cowl Point 200 1150Roof Height - 1600Roof Thickness - 20*Origin is ground plane at fore-aft location of AHP.

5

Optimization Design Variables

Vehicle Constraints

fixed dimensions based on,for example, carry-over

components or body features

Vehicle Parameters

Experimental Populations

combination of such factors as:- accommodation (e.g. seatand steering wheel placement)- comfort (e.g. headroom)- safety (e.g. steering wheelclearance and exterior vision)

Posturing Model

- preference for componentlocations as a function of body size- effects of restrictions due tocomponent locations (censoring)- random variance unrelated tobody and cab dimensions

Population Objective Functionboundary manikins (BM)

boundary manikins w/ random (BMR)

1000 driver population with variance (RS)

vehicle dimensions, adjustmentranges, component locations, etc.

factors such as overhall height, hoodstyling, maximum permissable interior

height, adjustment ranges, etc.

Population definitiongender mix, distributions ofstature, sitting height, etc.

Figure 3: Schematic of package optimization process.

where n is the total number of drivers in the population(n = 28 for BM, n = 280 for BMR, and n = 1000 for RS).

In each of the design scenarios described in the next sec-tion, the problem formulation is population-appropriate.When the boundary manikin populations are used (the aand b formulations), each of manikins must be accommo-dated, so FA = 1 is a constraint in the optimization. Afractional accommodation of 95% (FA ≥ 0.95), a typicaldesign target, was chosen for scenarios involving the en-tire RS population (the c formulation).

The optimization algorithm selects values for the designvariables (seat track location and adjustability) and the re-sulting package is scored using the product of the verti-cal and fore-aft adjustment ranges, which is approximatelythe side-view area of the adjustment range:

A = ∆x∆z. (3)

One measure of “cost” is the total amount of adjustabil-ity required to accommodate a desired fraction of drivers(e.g., 95% or FA = 0.95). At the expensive extreme is avehicle with ample adjustability that accommodates everydriver (FA = 1), no matter their size or behavioral prefer-ences. In contrast, a vehicle with no adjustability might beinexpensive to produce but the fraction of accommodateddrivers would be very low. The objective function for thecurrent analysis was to minimize the value of A, i.e., thesize of the seat track adjustment range, while maintain-ing accommodation for a specified fraction of drivers. Thegeneral optimization problem is

minimize f(x) = A

subject to 500 ≤ x ≤ 1000150 ≤ z ≤ 350

100 ≤ ∆x ≤ 3001 ≤ ∆z ≤ 100

{FA = 1 if BM;FA ≥ 0.95 if RS.

(4)

The simple bounds on x were selected so that the de-sign space encompassed by these values exceeds typi-cal design space given the fixed vehicle geometry in Ta-ble 4 and described below. This ensured that no variablewas at its limit when the algorithm converged to the op-timum. A global optimization algorithm, DIRECT (Jones,2001) was utilized for the examples, but the methodologyis not algorithm-specific. The maximum number of evalu-ations, which is a required input to the DIRECT algorithm,was arbitrarily chosen to be 5000. Additional constraintswere incorporated into the formulation as the scenariosdescribed below required them.

DESIGN PROBLEMS The optimization formulation,Eq 4, was solved under conditions defined by four designproblems. They were selected to show the utility of theapproach and to assess the suitability of the boundarymanikins to replace the much larger randomly sampledpopulation for design assessments.

There are three steps in each scenario, a, b, and c. Ina the problem is solved using the BM population andthe typical boundary manikin approach—residual posturevariance is not considered. The resulting design is thenevaluated for accommodation using the other two popula-tions, BMR and RS. Vertical disaccommodation is not pe-nalized, in keeping with current SAE practice (see J4004,SAE 2005). In b, the same design is evaluated, but avertical adjustment range of 40 mm is added. Vertical dis-accommodation is considered, and the resulting accom-modation for the BMR and RS populations is calculated.In c the optimization problem is solved again, but this timeusing the RS population. This gives an estimate of the

6

true magnitude of adjustability required to accommodate95% of the population under the specified conditions.

Scenario 1 Given a particular steering wheel position,find the adjustability requirements and location for the seattrack such that 95% of the driver population can sit intheir preferred location (not more than 5% are censoredon seat position).

Scenario 2 In the previous scenario, drivers whose de-sired posture was censored even a small amount wereconsidered disaccommodated. Permitting some smalllevel of disaccommodation, for example 10 mm, might al-low a cost-saving reduction in seat track adjustment rangewhile having a minimum impact on the overall acceptabil-ity of the design. It also better reflects the sensitivity ofa person to their seat position, since the “just noticeabledifference” for seat position is approximately 10 mm. Thisscenario explores the adjustability requirements and loca-tion for the seat track such that 95% of the driver popula-tion can sit within 10 mm of their preferred location. Thecalculation of NA was adjusted accordingly.

Scenario 3 The first two scenarios included a telescop-ing steering wheel, which may reduce the required rangeof seat adjustment. Scenario 3 quantifies that effect ofsteering wheel telescope by repeating Scenario 1 with anon-telescoping wheel (tilt only).

Scenario 4 Upvision and downvision requirements areincorporated into the design. This design was completedin two steps. First, the optimization procedure was usedto determined the location of the steering wheel pivot SWthat maximizes average downvision, DV , while ensuringthat either 100% (for BM population) or 95% (RS popula-tion) of drivers have at least 10 degrees of upvision. Theproblem was posed as:

minimize f(SW) = 1/DV

subject to 100 ≤ SWx ≤ 500400 ≤ SWz ≤ 700{FA = 1 if BM;FA ≥ 0.95 if RS.

(5)

where FA was calculated using the upvision criteriarather than seat location.

Once the steering wheel location is determined, we re-turned to the protocol used on the previous example.Eq. 4 was solved to determine the smallest seat track thatwill achieve 95% accommodation on seat position, upvi-sion, and maintaining the maximum average downvisionangle obtained earlier.

RESULTS

The results for each of the scenarios and conditions arereported in Table 5 and shown visually in Figure 4.

SCENARIO 1: SEAT TRACK LOCATION AND ADJUSTA-BILITY LIMITS Using the initial steering wheel design,the minimum seat track adjustment range to accommo-date all of the BM population was determined, using thefore-aft and vertical location of the seat track (x, z) and itshorizontal and vertical adjustment ranges (∆z, ∆z) as de-sign variables. The results are listed in Table 5, Row 1a.The linear seat track (no vertical adjustment) for the BMpopulation results because fore-aft and vertical seat posi-tion is predicted by linear functions that predict the mostlikely (mean expected) seat positions for people with thespecified body dimensions (Reed et al. 2002). The BMpopulation does not include the random variance in seatposition that is not associated with the vehicle and anthro-pometric descriptors.

The accommodation of the BMR and RS driver popu-lations were assessed in the design obtained with theBM population. Vertical disaccommodation was not con-sidered, mirroring current SAE procedures, which pre-dict only fore-aft accommodation (SAE 2005). Eighty-nine percent of BMR was accommodated within the spec-ified fore-aft range, and 96.3 percent of RS. For thisone-dimensional problem, accommodating the boundarymanikins, which nominally represent 95 percent of thepopulation, resulted in a slightly conservative design.

A better estimate of accommodation for BMR and RS canbe calculated by including disaccommodation on verticalseat position. A typical vertical adjustment range for apassenger car is 40 mm. Centering a 40-mm adjustmentband on the seat track length optimized for the BM popu-lation gives package 1b (Table 5, Row 1b) with an area ofadjustability of 7880 mm2. When both fore-aft and verticalaccommodation are considered, the seat track developedwith the BM population accommodates only 79 percent ofBMR and 81.5 percent of RS. Recall that a BM analysiswould indicate 95 percent accommodation in this design.

The seat location and adjustability required to achieve 95percent accommodation of RS was obtained through an-other optimization run with the same four design variables:x, z, ∆z, and ∆z (scenario 1c). The resulting seat ad-justment range is 207 mm long and 63 mm high, with arequired adjustability area of 13000 mm2. As expectedfrom the univariate optimization results, the seat track isonly slightly longer than that given by the optimization withthe BM population, but now the needed range of verticaladjustment is taken into account.

SCENARIO 2: MARGIN OF ACCOMMODATION If10 mm of seat-position disaccommodation is permitted,

7

Table 5: Optimization results.

Boundary Manikin with 1000 Driver PopulationScenario Boundary Manikin (BM) Random (BMR) with Variance (RS)

x z ∆x ∆z FA A FA x z ∆x ∆z A FA

seat track 1a 708 270 197 1 100 89 96.3seat track 1b 708 250 197 40 100 7880 79 7880 81.5seat track 1c 703 240 206 63 13000 95.0

margin 2a 711 270 184 1 100 90 96.9margin 2b 711 250 184 40 100 7360 89 7360 94.0margin 2c 708 251 201 38 7640 95.0

non-tele wheel 3a 711 281 197 1 100 89 96.0non-tele wheel 3b 711 261 197 40 100 7880 74 7880 77.2non-tele wheel 3c 688 252 249 61 15200 95.0

vision 4a 765 200 198 1 100 60 70.9vision 4b 765 180 198 40 100 7920 55 7920 62.2vision 4c 716 171 299 85 25360 95.0

x Fore-aft location of the full-down, full-rear corner of the seat adjustment range.z Vertical location of the full-down, full-rear corner of the seat adjustment range.∆x Fore-aft length of seat adjustment range.∆z Vertical height of seat adjustment range at each fore-aft position.A Sideview area of seat adjustment range.FA Fraction of population accommodated.

the required track length for BM decreases by 13 mm to185 mm. Neglecting vertical disaccommodation and in-cluding the 10-mm margin, this track would accommodate90 percent of BMR and 96.9 percent of RS. Adding 40 mmof vertical adjustment to this track length and position ac-commodates 89 percent of BMR and 94 percent of of RS,when vertical accommodation is considered. In the RSpopulation, 806 drivers were able to sit in their preferredlocation and an additional 134 were within 10 mm of it.These results are in Table 5, Rows 2a and 2b.

Rerunning the optimization with RS produced a track202 mm long and 38 mm high (Table 5, Row 2c) to accom-modate 95 percent of the target population with less than10 mm of seat-position censoring. Of the 950 drivers, 799were in their desired seat location and 151 were within10 mm of it. Comparing the results with scenario 1, per-mitting up to 10 mm of censoring allows 95 percent ofdrivers to be accommodated with a vertical seat track ad-justment range that is about 60% of the height and 5 mmshorter for a total adjustment area reduction of 41% (Fig-ure 4, Cells 1c and 2c).

SCENARIO 3: NON-TELESCOPING STEERINGWHEEL The steering wheel was allowed to pivot(as in previous examples), but not telescope (previousexamples had 50 mm of range). The same procedure fordetermining fractional accommodation was followed. Theresults are detailed Table 5, Row 3a.

Compared to Scenario 1, the adjustment range and hori-zontal location of the seat track are the same, but its ver-tical location is higher (Figure 4, Cells 1a and 3a). Ac-commodation levels are only slightly lower for a and b

(Table 5, Row 3a and 3b). When optimized for the RSpopulation, the required area, A is 17% larger, achievedthrough greater ∆x (Figure 4, Cells 1c and 3c). In otherwords, inclusion of a steering wheel with a 50 mm tele-scope range allowed a 17% reduction in the size of theseat track required to achieve the same level of driver ac-commodation.

SCENARIO 4: INCORPORATING UPVISION ANDDOWNVISION Using optimization to place the steeringwheel pivot location, SW, for a yielded (391, 518). Thislocation is rearward and downward of the original loca-tion. The maximum average downvision angle was 6.1degrees. The 280 BMR drivers, postured without restric-tion on seat position, had maximum average downvisionof 6.0 degrees. The 1000 RS drivers had maximum av-erage downvision of 6.1 degrees. The 280 BMR driverswith 40 mm of vertical adjustability had maximum averagedownvision of 6.0 degrees.

Using the full RS and optimizing for them achieves ac-commodation in both seat position and upvision whileslightly increasing maximum average downvision (to 6.9degrees). This is achieved by moving the steering wheelforward 27 mm, expanding the size of the seat track, andmoving it forward and down. To meet the 95% accom-modation criteria, all the drivers disaccommodated on up-vision must be accommodated on seat location. Conse-quently, the adjustment range is very large—effectively al-lowing all the drivers to sit where they would like (Figure 4,Cell 4c). Using a multiobjective approach and solving theoptimization problems simultaneously might reduce therequired size, depending on the relative weights given todisaccommodation on upvision and seat position.

8

A final analysis was performed that was similar to that inScenario 2. The full RS was used for an optimization inwhich drivers within 10 mm of their preferred seat loca-tion were considered accommodated. This reduced theoverall area of the adjustability range by 30%. The finaldesign is shown in Table 6 and, to facilitate comparisonwith the previous scenarios, in Table 5, Rows 4a-4c.

DISCUSSION

This paper presents a new approach to driver accom-modation modeling based on the application of optimiza-tion methods. The approach combines many of the ad-vantages of the current percentile-accommodation mod-els in SAE Recommended Practices with the flexibility ofhuman-figure-model approaches. The two most innova-tive aspects of the new approach are (1) the formulationof the packaging problem as an optimization, and (2) theinclusion of the variance in posture and other outcomemeasures (e.g., subjective assessments) that is not re-lated to vehicle factors or occupant descriptors. This vari-ance is included in percentile accommodation models butis commonly ignored in figure-model analyses.

The driver population used in this study, described by mil-itary data from ANSUR, is not representative of the typ-ical target population for a passenger car. However, themethods are independent of the choice of target popu-lation. Because of the large number of anthropometricmeasures that were included in the study, the ANSURdata are a good choice for algorithm development. Forapplication of this method, an appropriate design popula-tion should be identified for each new vehicle, including adetailed anthropometric description and the target gendermix. Relatively small changes in gender mix can changethe optimal package for a vehicle substantially.

Boundary manikin methods are widely used for figuremodel analyses in an attempt to capture anthropomet-ric variance in the target population without performinganalyses with hundreds of manikins. However, as thecurrent analysis demonstrates, boundary manikins oftendo not provide results comparable to those obtained withlarge, representatively sampled populations. As computerpower has improved, the rationale for using boundarymanikins has eroded, because it is now feasible to per-form most figure-model analyses programmatically. Usingrapid, deterministic posture prediction approaches, suchas the Cascade Prediction Model, thousands of driverscan be evaluated in a design in under a second of pro-cessing time on conventional desktop computers.

The boundary manikins used in this paper were sampledusing an approach that improves on the methods typi-cally used. In particular, the three anthropometric vari-ables that were selected were already known to be thosethat influence the outcome measures. In most boundarymanikin applications, body dimensions are included in theprincipal-component or factor analysis that affect only one

or a few of the possible outcome measures. Hence, thecurrent boundary manikin evaluation represents a best-case performance for the method; the results would oftenbe worse, relatively to the large randomly sampled pop-ulation, in typical usage. This is perhaps the most im-portant conclusion from the boundary manikin evaluation:the main drawback of using boundary manikins is that theactual accommodation remains unquantified, and may bequite different from that expected by accommodating all ofthe boundary manikins. The addition of random posturalvariance to the boundary manikins improved their abilityto represent the larger population. However, the problemof accurately estimating total population accommodationremains, because it is not clear how to cost the disaccom-modation of one or two boundary manikins.

The current SAE tools remain the industry standard forpackaging and provide well-validated results. The seat-ing accommodation model in J4004 and the eyellipse inJ941 have both been completely revised in the last fiveyears and can be used effectively for most vehicles. How-ever, these tools do not consider some important adjust-ment features in newer vehicles, such as adjustable ped-als. They are also not well suited to quantitative tradeoffstudies, in which, for example, the relative costs of addingadjustability to the seat and steering wheel are consid-ered. By design, the current methods can reproduce theseat position and eye location distributions that underliethe J941 and J4004 models when the appropriate con-straints are applied (i.e., fixed steering wheel position, nocosting of vertical seat accommodation, and no posturaleffects of headroom or vision). However, the optimiza-tion formulation and the ability to experiment using a largenumber of virtual drivers makes the new approach muchmore flexible.

The new method has broad applicability to vehicle designproblems. The central tenet of this approach is that vari-ance in outcome measures, whether postural or subjec-tive, should be partitioned among vehicle factors, occu-pant factors, and the residual not accounted for by theother two categories. The corresponding primary limita-tion of the method is that it can only be used when goodestimates of variance in these three categories are avail-able. Developing these estimates usually requires rela-tively large, carefully designed studies of occupant behav-ior and subjective response. The scope of work requiredto develop appropriate models often causes manufactur-ers to short-circuit the process by relying on heuristics andbenchmarking, yet the cost of remedying errors that wouldhave been prevented by well-developed models is oftenan order of magnitude greater than the cost of gatheringthe needed data.

Another strong advantage of the new approach is that theanalysis tools are constructed in the context of an opti-mization framework. The process of repackaging a ve-hicle when constraints are changed often takes too longfor packaging issues to be taken into account in dynamic

9

Table 6: Summary of RS results from upvision and downvision scenarios utilizing three approaches.

SWx SWz x z ∆x ∆z A DV FA(mm) (mm) (mm) (mm) (mm) (mm) (mm2) degrees %

traditional 391 518 765 200 199 not considered n/a 6.1 70.9traditional +40 mm 391 518 765 180 199 40 7920 6.1 62.2

strict accommodation 364 520 716 171 300 85 25360 6.9 95.0marginal (10 mm) accommodation 364 520 727 180 280 64 18160 6.9 95.0

trade-off studies. Implementations of the SAE packag-ing models as macros in computer software has reducedthese turnaround times, but missing from those systemsis the ability to optimize the design in the face of multipleconstraints and objectives. In any case, the optimizationcan only be performed when the appropriate cost func-tions, derived from human-subject studies, are available.The next steps in the development of this new methodol-ogy will be case studies that explore the range of potentialapplications, incorporating factors such as crash safetyand financial cost, and the conduct of new experimentsto increase the number of subjective cost functions thatare available.

ACKNOWLEDGMENTS

The driver-posture research incorporated into this newpackaging methodology was conducted over the past 15years at the University of Michigan Transportation Re-search Institute with sponsorship from many companiesand organizations, including the Motor Vehicle Manufac-turers Association, the Association of Automobile Man-ufacturers, the Alliance of Automobile Manufacturers,BMW, Chrysler, Ford, General Motors, Johnson Controls,Lear Corporation, Mazda, Peugeot-Citroen, and Toyota.The initial investigation of a driver posture implementationin an optimization framework was supported by the Au-tomotive Research Center at the University of Michigan.Subsequent work has also been supported by The Penn-sylvania State University.

REFERENCES

Bittner, A. C. (2000). A-CADRE: Advanced family ofmanikins for workstation design. Proceedings of theIEA/HFES 2000 Congress. Human Factors and Er-gonomics Society, Santa Monica, CA.

Chaffin, D.B. (2001). Digital Human Modeling for Vehicleand Workplace Design. Society of Automotive Engineers,Warrendale, PA.

Dempster, W.T. (1955). Space requirements of the seatedoperator: Geometrical, kinematic, and mechanical as-pects of the body with special reference to the limbs.WADC Technical Report 55-159. Wright-Patterson AFB,OH: Wright Air Development Center.

Flannagan, C.A.C., Manary, M.A., Schneider, L.W., andReed, M.P. (1998). An Improved Seating Accommoda-

tion Model with Application to Different User Populations.Technical Paper 980651. SAE Transactions: Journal ofPassenger Cars, Vol. 107.

Geoffrey, S.P. (1961). A 2-D mannikin–The inside story.X-rays used to determine a new standard for a basic de-sign tool. Technical Paper 267A. SAE International, War-rendale, PA.

Gordon, C. C., Churchill, T., Clauser, C. E., Bradtmiller,B., McConville, J. T., Tebbetts, I., and Walker, R. A.(1989). 1988 Anthropometric Survey of U.S. Army Per-sonnel: Methods and Summary Statistics. Final Report.(NATICK/TR-89/027). U.S. Army Natick Research, Devel-opment and Engineering Center, Natick, MA.

HFES 300 Committee (2004). Guidelines for Using An-thropometric Data in Product Design. Human Factors andErgonomics Society, Santa Monica, CA.

Jones, D.R. (2001). The DIRECT global optimization al-gorithm. In Encyclopedia of Optimization, pp. 431-440.Kluwer, Norwell, MA.

Loczi, J. (2000). Ergonomics program at Freightliner.Technical Paper 2000-01-3402. SAE International, War-rendale, PA.

Manary, M.A., Flannagan, C.A.C., Reed, M.P., andSchneider, L.W. (1998). Development of an ImprovedDriver Eye Position Model. Technical Paper 980856. SAETransactions: Journal of Passenger Cars, Vol. 107.

Manary, M.A., Flannagan, C.A.C., Reed, M.P., andSchneider, L.W. (1999). Human subject testing in supportof ASPECT. Technical Paper 990960. SAE Transactions:Journal of Passenger Cars, Vol. 108.

Meldrum, J.F. (1965). Automobile driver eye position.Technical Paper 650464. SAE International, Warrendale,PA.

Parkinson, M.B., Reed, M.P., Kokkolaras, M., and Pa-palambros, P.Y. (2005). Robust truck cabin layout opti-mization using advanced driver variance models. Pro-ceedings of 2005 ASME Design Engineering TechnicalConference. ASME, New York.

Porter, J.M., Case, K., Freer, M.T., and Bonney, M.C.(1993). Computer-aided ergonomics design of automo-

10

biles. In Automotive Ergonomics, ed. B. Peacock and W.Karwowski. Taylor and Francis, London.

Reed, M.P. and Flannagan, C.A.C. (2000). Anthropo-metric and postural variability: limitation of the bound-ary manikin approach. Technical Paper 2000-01-2172.SAE Transactions: Journal of Passenger Cars Mechani-cal Systems, Vol. 109.

Reed, M.P., Manary, M.A., Flannagan, C.A.C., andSchneider, L.W. (2002). A statistical method for predict-ing automobile driving posture. Human Factors. 44 (4):557-568.

Reed, M.P., Roe, R.W, and Schneider, L.W. (1999). De-sign and development of the ASPECT manikin. TechnicalPaper 990963. SAE Transactions: Journal of PassengerCars, Vol. 108.

Roe, R.W. (1993). Occupant Packaging. In AutomotiveErgonomics, ed. B. Peacock and W. Karwowski. Taylorand Francis, London.

Seidl, A. (1994). Das Menschmodell RAMSIS: Analyse,Synthese und Simulation dreidimensionaler Korperhaltun-gen des Menschen [The man-model RAMSIS: Analysis,synthesis, and simulation of three-dimensional humanbody postures]. Ph.D. dissertation, Technical Universityof Munich, Germany.

Society of Automotive Engineers (2005). SAE Handbook.SAE International, Warrendale, PA.

11

270

708

197

100

250

708

197

81.540

240

703

206

95.063

251

708

201

95.038

250

711

184

94.040270

711

184

100

261

711

197

100

261

711

197

77.240

252

688

249

95.061

171

716299

95.0

85180

765198

62.2

40200

765

198

100

z∆x

FA

x

∆z seat trackadjustment range

preferred seat locationfor one driver

(a) traditional boundarymanikin approach with

BM population

(b) RS population in seat with40 mm vertical travel

located by BM

(c) RS population in seatoptimized for 95%accommodation

(1)

sim

ple

seat

trac

k pl

acem

ent

(2)

mar

gina

lac

com

mod

atio

n(3

) no

n-te

lesc

opin

gst

eeri

ng w

heel

(4)

visi

onco

nstr

atin

s

Figure 4: Seat track location and adjustment range results from the four scenarios.

12