Three Takeaways from NTSB’s Preliminary Report on the First Fatal Automated Vehicle Crash
May 31, 2018
Two months after the fatal crash in Tempe, Arizona involving an Uber test vehicle in self-driving mode, the National Transportation Safety Board (NTSB) released a preliminary report on the crash.
There was one operator in the vehicle when it struck a pedestrian, Elaine Herzberg, walking a bicycle. The crash occurred at 9:58 pm on a four-lane road. The three and a half page NTSB report leaves many questions unanswered, but it is the first official federal report of a crash involving an automated driving system, so it sets the precedent for future reports regarding incidents involving highly automated vehicles (HAVs). It includes information about the vehicles, vehicle operator, pedestrian, road design, and circumstances around associated with the event.
On March 19 (the day after the crash) NTSB sent a team to Tempe to examine the crash. The agency“investigates select highway crashes that can advance knowledge of broad or new safety issues.” The Tempe crash allows NTSB to investigate vehicles with automated driving systems interacting with the environment, other vehicles, and vulnerable road users.
Three main takeaways from the preliminary report:
The report highlights the problems with “false positives”.
NTSB’s report notes that emergency braking was not engaged at the time, raising questions around Uber’s decision in disabling certain safety functions such as the driver alertness detector, road sign information detection, and emergency braking while the vehicle is not in manual mode. The latter, Uber states, is to create smoother, less erratic driving patterns.
The report does not clearly state the reason for the crash, but the lack of the vehicle’s decision to maneuver or brake seems to be a combination of the inability of the vehicle to correctly identify the pedestrians and estimate her trajectory, and the disengagement of the emergency braking system. By first identifying Ms. Herzberg as a vehicle, and then as a cyclist (note that NTSB uses the word “bicycle” instead of “cyclist,” which does not imply an average speed or acceleration profile), the vehicle did not brake in time (though it did identify the need for emergency braking), as neither the technology nor the operator identified the person correctly in enough time to maneuver or stop.
Safety drivers must be alert at all times to ensure safe operation of the automated system.
The report does not explain the lack of a driver alertness detection system at all times, but it does state that, “According to Uber, the developmental self-driving system relies on an attentive operator to intervene if the system fails to perform appropriately during testing.” This statement suggests that the driver alertness detector should in fact be engaged at all times.
Overall there is no reason not to maximize communication between the vehicle, the vehicle operator, and other road users. Visual, tactile, and audio recognition and communication could be clearer in all directions for the automated systems to monitor and communicate with the operator, for HAVs to recognize and communicate with other road users, and for other road users to recognize HAVs. Many forms of communication could be lost when there is no longer a human operator visible in the vehicle, namely eye contact, thus methods such V2X communication, honking, lights, visible identifying features, could improve road safety.
NTSB places significant blame on the pedestrian
The information presented, however, skews towards what pedestrian advocates often refer to as pedestrian victim blaming. For example, the description of the pedestrian in the report notes that she was wearing dark clothing, but does not likewise comment on the color of the vehicle, which could also have played a role in the crash. NTSB further commented that the bicycle being walked by Ms. Herzberg did not have side reflectors. All of these details are important to understand what types of surfaces cannot be recognized and identified in a timely and accurate manner by the self-driving technologies. Suggesting that vulnerable road users should dress a certain way to avoid fatal collisions is not an equitable approach to improving road safety.
As an independent agency with the goal of “promoting a higher level of safety in the transportation system” the NTSB should be aware of the implications of what is and is not reported to avoid undertones of causation in reports that “by nature [do] not contain probable cause.” Hopefully the final report on the Tempe crash will contain much of the missing information on the functionality of the technologies used in the vehicle and present a more balanced discussion of the road users.