Robots Cars vs. Human Drivers: Who’s at Fault?
Dangerous terribly bloody motorcrash, destructive motorcrash. — The Sugarcubes
I will get down rollin’ tonight/Nobody’s fault — Led Zeppelin
If you’ve listened to or read any of the presentations I’ve made on connected and automated vehicles over the past two years, you’ve heard me stress the following concerns: First, collisions involving automated vehicles (AVs) will most likely be caused by the actions of other human-driven vehicles. Second, in states like California, proportional liability laws will mean that smart lawyers will look to the agencies that build, operate, and maintain the roadways when collisions between AVs and human-driven vehicles occur. The first concern has already been validated. The second concern has proven elusive, until recently.
As reported in Popular Mechanics magazine in a recent article, an AV collided with a motorcycle in San Francisco on December 7, 2017. This day might live in infamy for a new reason, as this might be the first day that people realized the broader implications of AVs in terms of mobility in cities.
The article talks about a “blame game” ensuing between the motorcycle driver and General Motors, precipitated by a lawsuit brought by the driver of the motorcycle. From GM’s point-of-view, the driver of the motorcycle is at-fault for illegally attempting to pass on the right. This position seems to be justified by the report issued by the California Department of Motor Vehicles. In the report, the DMV writes that “the motorcyclist was determined to be at fault… in violation of CVC 21755(a).” When you read the California Vehicle Code section 21755(a), it is clear that the motorcyclist should only have attempted to pass the GM AV on the right when “conditions permitting that movement” ensured the safety of the motorcycle driver.
It will be very interesting to monitor how this case continues, because when I read the DMV report, I notice something very glaring. The report describes the events preceding the collision as follows:
“The Cruise AV was traveling in the center of three one-way lanes. Identifying a space between two vehicles (a minivan in front and a sedan behind) in the left lane, the Cruise AV began to merge into that lane. At the same time, the minivan decelerated. Sensing that its gap was closing, the Cruise AV stopped making its lane change and returned fully to the center lane.”
Did you catch that? Do you notice anything missing from the report? Maybe not. Let me explain.
The California Vehicle Code provides very specific rules for vehicles traveling on one-way roads, and very specific rules for changing lanes on one-way roads. According to CVC 21658(a), a vehicle “shall not be moved from the lane until such movement can be made with reasonable safety.” The CVC also provides the specific steps that must occur when a vehicle changes lanes. This is spelled out in CVC 22107, which says: “No person shall turn a vehicle from a direct course or move right or left upon a roadway until such movement can be made with reasonable safety and then only after the giving of an appropriate signal in the manner provided in this chapter in the event any other vehicle may be affected by the movement.”
Coming back to the collision between the AV and the motorcycle, two things jump out to me:
- The DMV report doesn’t say anything about the GM AV using a turn signal before attempting to move from the center lane to the left lane. This could be an oversight on the part of the report, though I wonder with true machine learning, are the AVs learning that from bad behaviors exemplified by human drivers not using turn signals regularly?
- The DMV report says that the GM AV “identified a space between two vehicles.” The report doesn’t say if the GM AV adequately determined if that movement could be made safely. The fact that the vehicle was estimated to be travelling at 12 mph means that not much additional space was necessary for braking. At the same time, the fact that the DMV report identifies the deceleration of the minivan in the left lane as the reason the GM AV aborted its attempted lane change raises some concerns about the vehicle’s ability to determine what is and what is not a safe maneuver.
The two points above will be crucial, in my opinion, when digging deeper into this collision and its causes. If the GM AV determined that the lane change maneuver could be executed “with reasonable safety,” and was travelling at a speed of only 12 mph, then why was the sudden braking of a minivan so disruptive to the maneuver? It seems that the algorithms used to assess the reasonable safety levels of maneuvers need to be, at a minimum, evaluated. The question of whether the AV used a turn signal is also important, as it will inform the broad process used by the AV to evaluate its surroundings, make decisions, and execute maneuvers.
Last, it will be very enlightening to see if any of the various lawyers assigned to the suit start looking at the quality of the supportive roadway infrastructure. Were the lane stripes visible and up-to-date? Were the street signs clear? Why did the GM AV decide that it needed to be in the left-lane anyway, and how much of that decision was guided by traffic control devices installed and maintained by the City of San Francisco? To me, this will be the more important element of the case, as it would have far-reaching implications for any public agency looking at AVs.
Buckle your seatbelts, keep your hands and feet inside the vehicle, and secure all personal belongings, because this ride has only just begun.
