Vigilance of truck drivers

1 The most important factor in the experimental design is driving schedule. Four driving schedules were used. Some aspects of these are shown in Table 1.

Table 1. Design Driving Schedules

2It will be seen that 20 drivers served in each of the schedules. Five instrumented tractor traders were used for each driving schedule, and 5 drivers per week made 4 or 5 consecutive trips. All drivers were regularly-employed drivers hauling revenue freight over the motor carrier’s regular route. It will be seen from the second column of Table 1 that the schedules involved either 10 hours of driving or 13. The 10-hour data collections were done in the U.S., where hours-of-service rules prohibit driving more than 10 hours without an 8-hour break. The 10-hour drives were all from St. Louis, Missouri to Kansas City, Missouri and return. This trip takes on average slightly less than 10 hours of driving, plus break-time and time to switch traders and do performance tests at the turnaround point. Two types of 10-hour drives were employed : the « daytime departures » schedule, and the « rotating shift » schedule. The 10-hour daytime condition was intended as a baseline or control condition representing what we judge to be the least stressful way to reach the U.S. hours-of-service limits, namely morning departures at the same time each day for 5 consecutive days.

3 The second driving schedule, the « 10-hour rotating shift » schedule, employed another group of 20 drivers (and a different motor carrier) over the same route from St. Louis to Kansas City and back. However, the schedule was intentionally made more demanding to reflect the reality that drivers, particularly when going across country, often operate on a drive-rest-drive cycle of less than 24 hours. Because of the 80 minutes of performance testing that we required of the drivers each day, plus unavoidable delays experienced in the truck terminals, we found that a 21-hour cycle was the shortest we could achieve. Thus, the drivers on this schedule left 3 hours earlier each day. All the drivers left in the morning of the first day, and by the fifth day all the drivers were leaving 12 hours earlier, in the evening, and driving through what probably was the depths of their circadian cycles. The results in terms of the performance, physiological, and subjective data should be very interesting.

4The third and fourth driving schedules were conducted in Canada, which permits up to 13 hours of driving before requiring 8 hours off duty. The « night departure » schedule involved 20 drivers in tractor-trailer drives from Toronto to Montreal and return. The « day departure » drives were performed with another group of 20 drivers going from Montreal to Toronto and return. Because of the greater number of hours per trip, compared to the U.S. schedules, the Canadian drivers reached another hours-of-service constraint, namely the 60-hours-in-7-days rule, in 4 days rather than 5.

5The « raw » data from all four driving schedules are summarized in Table 2.

6The 465 trips produced two video tapes per trip (outbound and inbound), for a total of 930 video tapes representing about 5,000 hours of combined images of driver’s faces and the roadway. Most of the sensors in the truck (steering wheel angle, lane tracker, etc.) were recorded by the instrumentation computer, which generally produced about one diskette of compressed data for the 10-hour trips and one and a half diskettes worth for the 13-hour trips. Thus, there are about 500 diskettes containing about 700 megabytes of raw data. The physiological recordings consist primarily of EEG tapes made by Oxford Medilog ambulatory recorders which were then transcribed by Oxford equipment onto optical disks. This produced a total of twenty-eight 800 megabyte optical disks representing 10,000 hours of EEG recordings. Various questionnaires and subjective rating scales were also employed, which resulted in about 1000 sheets of paper.

7The raw data are too voluminous to be easily manipulated or readily summarized. We are in the process of reducing the raw data. Some features of the reduced data base are shown in Table 3. The first row of this table indicates that we are sampling 1 minute of video out of every 30 minutes to prepare a Borland Paradox « video details » table. Each 1 -minute sample produces a summary record with 50 fields indicating traffic and weather conditions, driving lane or lanes used, any apparent lane drift (this is also measured quantitatively with a lane tracking camera), appearance of drowsiness, yawns, droopy eyelids, slow blinks, head nodding, etc. Any apparent drowsiness will be cause for closer scrutiny of the video surrounding the event, so the sampling rate for events of interest will be much higher than one every 30 minutes. Naturally, any apparent drowsiness will also cause us to pay close attention to the performance measures and physiological measures as well. The video, computer, and EEG recorder time bases are synchronized to a precision of finer than one second to facilitate such comparisons. Any episodes of drowsiness or other events of a critical or interesting nature that we uncover will be clearly flagged in the data base, so that other investigators may also study them.

8The second row of Table 3 refers to the « edited » truck computer data base. The data (for example, lane tracker data) will be edited both automatically and manually. Examples of automatic editing include « killing » data points while left or right turn signals are on or when truck speed is below a threshold value. Manual edits will also be made by using an interactive time series presentation which allows manual « killing » of, e.g., lane tracker data affected by poor lane delineation, off-ramps, on-ramps, etc. It is important to note that the « killing » process does not remove any of the original raw data, but simply sets various flags in the data base to indicate that the data points should not be used in calculating descriptive statistics. Thus, as may be seen in Table 3, the size of the edited data base is approximately the same as the raw data base, namely 700 megabytes on one optical disk.

9After editing the truck computer data base, the data will be reduced by aggregating them into one-minute epochs. This is indicated in the third row of Table 3. Summary statistics are calculated for each one-minute epoch to prepare a Paradox table that will be comprised of about 335,000 records of 100 fields each. This will produce approximately 335 megabytes of data. This Paradox table will permit easy manipulation of the data to make comparisons and do statistical tests among various conditions represented within the data base.

10The sleep data are being subjected to manual scoring according to the method of Rechtshaffen and Kales. The resulting analysis is encoded to provide a means for evaluating sleep quality during each principal sleep period. This is a very labor intensive effort, but it results in an enormous compression of the data, from 22,400 megabytes down to 14 megabytes. The waking physiological data, primarily EEG data during driving, will be subjected to spectral analysis to attempt detection of critical events such as drowsiness. Also, any episodes of napping will be manually scored in the conventional manner. (The drivers were encouraged to pull off the road and nap any time they felt that would be useful ; this occurred on several occasions.) The results of these events of interest will be tabulated in a Paradox table. The size of that table will depend on the number of « critical events » that we eventually detect.

11Finally, the various questionnaires and rating scales will be keyed to disk and represented in Paradox tables comprising approximately four megabytes to facilitate their integration with the rest of the data base.

12The entire reduced data base should fit on two or possibly three 800-megabyte optical disks, which will be made available to interested researchers for the cost of the media and the labor of copying the data. The data should be available about mid- 1994. We at Essex will be analyzing the data ourselves as part of our current effort, but our experience has been that one team of investigators rarely is able to extract all the useful information from a given data set. Far too often, data that have been painstakingly collected wind up sitting on a shelf somewhere until the recording medium is corrupted or there are no longer devices to read it with. We intend to archive these data on long-life optical disks and to promote use of the data by other researchers. This data collection has been a very substantial effort for us, and we do believe the data base will yield knowledge leading to safer and more efficient highways. We hope you will share in searching for that knowledge. If you have an interest in using the data, please let me know.