Abstract:The original proposal to fund this research was based upon a type of comparable systems analysis, seeking to use rail transit service safety analysis as comparable systems. A literature search at the ITS library in UC Berkeley could find no useful references for comparable rail systems. At this point, the research team consulted with PATH staff to identify a more suitable and productive research scope.

The automated highway scenario considered in this research is one in which the freeway is partially automated, that is, only the left or median lane is automated. The analysis is concerned with accidents that originate outside the left lane and result in a vehicle or debris being deposited in the left lane or on the left hand side median. These accidents are of interest because they can clearly affect the safety of the automated lane, yet are not part of the automated control system itself.

The specific goal of this research is to begin to determine the scope of the safety problem posed by accidents of this type. The necessary objectives in such an investigation include:

• 1. To develop a method to identify the relevant accidents from the population of reported accidents on Califomia freeways.
• 2. To test the accuracy of the method by comparing relevant accidents identified by a computer search with those identified by manual review of paper accident reports.
• 3. To use the method on a test section of California freeway. Compare the characteristics of the relevant crashes to those of the broader accident population.
• 4. To assess the implications of the accident comparisons for automated highway system design, at least at the concept stage.

As a test case this study uses accident data for I-10, the Santa Monica Freeway, between I-405 and I-110 in California in the years 1986 and 1987. The method to identify relevant accidents was found to be sound and a cross-classification analysis was performed on the accidents selected to identify the factors contributing to the accidents of interest. A spatial analysis was conducted to help determine if relevant accidents were clustered near on and off ramps, an issue with important implications for automated highway design.

The analysis revealed a total of 2069 accidents on the section of the I-10 freeway of which 273 were relevant. A subsample of accidents, selected randomly, revealed that the method was accurate in identifying relevant accidents with 95% confidence. Relevant accident attributes were compared to the attributes of all other accidents in a series of cross classification tables. The comparisons revealed that the accident status was independent of the weather and the month during which the accident occurred, but dependent on the type of collision, time of day, lighting conditions, road surface condition, number of vehicles involved, move preceding collision, primary collision factor, and sobriety / drug / physical condition of driver.