• Federal Highway Administration >
• Publications >
• Research Publications >
• 08019 >
• Sect5.Cfm >
• Development of a Driver Vehicle Module (DVM) for the Interactive Highway Safety Design Model (IHSDM)

Development of a Driver Vehicle Module (DVM) for the Interactive Highway Safety Design Model (IHSDM)

PDF files can be viewed with the Acrobat® Reader®

SECTION 5. SUMMARY AND CONCLUSIONS

Key DVM Application Constraints

Application of the DVM is bounded by a number of constraints associated with the conceptualization and implementation of the model. These constraints include:

• The driver is experienced with the driving task in general.
• The driver makes appropriate decisions and control actions when given good perceptual information.
• The highway driving situations are typical and relatively relaxed.
• The vehicle performs properly.
• The driving task is basically limited to the operational task of regulating vehicle path and speed.

Additional Model Enhancements

Work on this project has revealed a number of areas in which the DVM could benefit from further development. The current DVM implementation would need to be modified to treat the highway conditions and/or driver behaviors discussed below.

Cruise Control and Compound Curves

The request for information on potential model enhancements has arisen largely from an inquiry concerning the potential application of the DVM to a segment of Massachusetts Interstate 95 containing a compound curve in one direction followed by a curve in the opposite direction. The compound curve consists of a lead-in curve, a central (sharper) curve of lower radius, and a lead-out curve having the same radius as the lead-in curve. The inquirer was concerned about the potential for rollover where the horizontal alignment reverses, particularly at times when speeds in excess of 80 mi/h are routine.

We deal first with the issue of speed. We assume that the concern is for drivers who maintain 80+ mi/h throughout the curves. The present implementation does not have the capability to impose this condition in a credible manner. The existing implementation does allow the user to specify a very large free speed and to assume that speed limits are ignored, but even under these assumptions the DVM would slow down for curves. In principle one could force a constant speed by specifying zero SD (the current implementation does not allow a SD less than 100 m), but then how would the driver be able to steer?

A developmental version of the DVM has been created for the purposes of calibrating vehicle lateral and longitudinal response which allows the user to specify a fixed throttle position. In this configuration, the driver continues to steer the vehicle but does not control throttle or brake. If desired, this capability could be included in the public-release version of the IHSDM.

The assumption of a fixed throttle position would not be reasonable for driving over a typical two-lane rural road with segments of varying horizontal curvatures. It might be more reasonable for highways designed to interstate standards in which curved segments are required to have relatively large radii of curvature. Even so, a fixed throttle does not guarantee fixed speed because of the accelerating and decelerating effects of down slopes and up slopes, respectively.

Incorporation of a submodel for cruise control would allow a more credible representative of actual driver behavior. In this configuration, the free speed parameter would be superseded by a minimum speed parameter, and the speed-control component of the DVM would be modified to regulate speed about this minimum speed using throttle only. Such an implementation would allow the vehicle to proceed faster than the desired speed on steep down slopes, but such behavior is representative of driving with cruise control.

Implementation of a cruise-control option in which a constant desired speed is specified for the entire run should require only a modest software development effort. Implementation of the capability for the driver to transition from cruise control to driver control of speed is not recommended at this time because of the absence of data for determining the rules for transition. Accordingly, implementation of fixed cruise control is recommended only for highways having consistent horizontal geometry; that is, situations in which a driver might be expected to leave the cruise control setting untouched over the roadway of interest.

The DVM does not properly treat speed decision in compound curves that consist of three consecutive segments of constant curvature; this condition has not yet been addressed. One remedy would be to augment the DVM to recognize such compound curves and allow the driver to cut only the central curve.

Driver Behavior on Short Tangents

Further experimental and theoretical studies are recommended for developing a general model for the speeds in tangents connecting horizontal curves. Such a model would allow the application of a model (for which data currently exist) for predicting accelerations and decelerations as functions of predicted speed changes.

Horizontal Sight-Distance Limitations

Analysis of driver behavior observed in the VTTI on-road study suggested that horizontal sight-distance limitations may have influenced vehicle speed on tangents. An experimental study of on-road or in-simulator driver behavior accompanied by model analysis is suggested to improve the capability of the DVM to model these effects.

More Flexible Model for Curve Cutting

The current model for curve cutting is limited to cutting to the inside of the curve and is applicable only to curves of constant radius. Consideration of the trailer wheels as well as the cab wheels would provide a basis for allowing trucks to track to the outside of a curve. Extension to compound curves would likely require a substantial modification of the model for curve cutting.

Effects of Driver Eye Height and Grade Differences on Curvature Estimation

The perspective view of a horizontal curve is influenced both by the height of the driver's eye above the road and difference in grade between the approach tangent and the curve. A study of on-road and/or in-simulator driver behavior is suggested to quantify the extent to which such perceptual differences influence the manner in which drivers approach and negotiate curves, accompanied by model development to adequately reflect such effects in the DVM.

Driver Expectation

The element of surprise cannot be programmed into the DVM. That is, violation of driver expectancy per se is not something that the DVM will flag. Therefore, the flags that were generated in testing the scenarios were potentially influenced more by the characteristics that make up an aggressive driver (e.g., waiting until the last possible moment to decelerate, driving fast through curves) than by the alignment itself. Future data collection and modeling efforts for the DVM should seek to add a parameter that allows driver expectation to be varied.

Additional User Interface Enhancements

Because the enhancements below were not deemed critical to the central goal of developing the DVM, implementation of these enhancements was not undertaken and is recommended for future model development.

Enhance the DVM Output Information so that it Better Conforms to End-user Needs.

DVM users have suggested a number of improvements in output presentation and format. Specifications for realizing these improvements are summarized in Table 13 . Accompanying each specification are suggestions related to its implementation; these suggestions are based on comments provided by the participants in a URA. Each of these specifications may be implemented by the FHWA as they deem appropriate.

Develop new Measures of Effectiveness (MOEs) Based on Degree of Speed Change and Available SD.

Additional MOEs were suggested by the respondents to the URA conducted in task A.2. Two of these suggestions are addressed here: providing alert levels related to available SD and to speed changes associated with horizontal curves.

As discussed below, the DVM currently includes alerts for critical variables that are computed from ensemble statistics obtained from multiple trials. These alerts are associated with predicted probabilities of exceeding some criterion value, where only a single criterion value is associated with a particular variable (e.g., the probability that the vehicle lateral path exceeds the lane boundary).

To be consistent with the treatment of speed changes used in the DCM, the proposed alerts for speed change will involve two criterion values that define three ranges of predicted speed differences. In this case, the philosophy of predicting the probability of an out-of-bounds situation does not readily apply, and we introduce the notion of basing alert levels on the results of a single simulation trial or the ensemble mean of multiple trials. As we show below, alerts for some of the critical variables can be defined for both deterministic and statistical analysis.

Statistical Alerts

Statistical alerts are currently provided for path error, X and Y skid indices, and rollover index. Computation of an additional statistical alert for SD is proposed.

The SD requirement predicted by the DVM will in general differ from SD requirement specified by the "Green Book^"(8) because the DVM is a dynamic model that predicts an instantaneous vehicle speed that in general is influenced both by highway geometry and assumed driver characteristics. The values for velocity used in computing the required SD in the "Green Book"⁽⁸⁾ are based on the assumed design speed of the highway-a static variable.

The suggested procedure for computing SD alert levels is as follows:

1. Perform a multiple-trial simulation in which representative values are assigned to the noise processes associated with the perception of variables relevant to vehicle control. A simulation involving 30 to 40 trials is recommended for stable statistical results. For each trial, record the time histories of the available SD (currently computed in the DVM) and the required SD.
2. Upon the termination of the simulation trials, perform an ensemble analysis of the recording of instantaneous SD to compute, at regular station intervals, mean and standard deviation of available SD. Also compute the ensemble mean of the required SD.
3. On the assumption that the available SD has a Gaussian distribution, compute the probability that the available SD exceeds the instantaneous required SD.
4. Define an appropriate alert level for the station at which the following transitions occur:
   • Red alert if the probability of insufficient SD equals or exceeds 1%.
   • Yellow alert if the probability transitions into the range of greater 0.1 % but less than 1%.
   • Green alert if the probability falls below 0.1%.

These probability criteria are not theoretically based but were selected by the developers to provide a framework for conveying relative risk.

The SD alert levels can be included as an additional column in the tabular presentation of alert levels as shown in Table 14 for the four current levels. The max alert level should then indicate the maximum alert level of the five component alerts.

Deterministic Alerts

The DCM defines the following alert levels for speed decreases (in meters) associated with horizontal curves:

Green alert: V_diff £ 10.0

Yellow alert: 10.0 < V_diff £ 20.0

Red alert: 20.0 < V_diff

where

V_diff is the speed reduction.

The same definitions are recommended for the DVM.

The table of deterministic alerts should also include alerts related to lane position and SD. For the lane position alert, the instantaneous lane position is tested against the criterion value (default is lane boundary), a red alert is associated with transitioning outside the criterion value, and a green alert for transitioning into the lane. The instantaneous lane position is used in the calculation for a single-trial simulation; the ensemble mean of lane position is used for a multi-trial simulation. Similarly, red and green alerts are associated with the SD becoming insufficient or sufficient. Deterministic yellow alerts are not defined for either lane position or SD.

Add the Ability to Compare Time Histories from Multiple Model Runs on the Same Graph.

At present, time histories from multiple model runs can be compared on the same graph only by using software (such as Excel) that operate on the output data files created by the DVM. Over half the participants in the URA conducted in task A.2 indicated they would have a use for being able to make such comparisons within the DVM environment. Table 15 lists specifications for providing comparisons of the output of multiple model runs from within the IHSDM environment. Comparison Mode Specifications lists the recommended format of the comparison outputs. Recommended Features Specifications describes some desirable features that may be incorporated to improve the utility of the output.

Recommendations for Using the DVM

Despite the constraints and limitations associated with the current version of the DVM, the DVM does indeed provide some real value to highway designers. If its applicability can be expanded to other roadway types (e.g., nonrural roadways) and driving conditions (e.g., multiple vehicles on the roadway, bad weather), it can become more broadly valuable. The DVM has already been used by a small number of highway designers to evaluate new rural road designs and has produced some useful and interesting results. Given its limitations, it is perhaps most useful (in its current form) as a tool for identifying those portions of a candidate roadway that are clearly unsafe and should perhaps be re-designed or subjected to further analysis.

Future R&D Recommendations for the DVM

Key recommendations for future DVM R&D are listed above. Additional enhancements to the DVM that we recommend include the addition of:

• Other vehicles into the driving scenario.
• Multi-lane highways.
• Traffic signals.
• Intersections.
• Multiple driver tasks.

A key recommended enhancement to the DVM is a full Java implementation that eliminates Visual Basic components.

Previous | Table of Contents | Next