The data indicates a decreasing incidence of disengagements over the time period covered by the report, Sept. 2014 to Nov. 2015.
According to the report the DMV defines disengagements as deactivations of the autonomous mode in two situations: (1) “when a failure of the autonomous technology is detected,” or (2) “when the safe operation of the vehicle requires that the autonomous vehicle test driver disengage the autonomous mode and take immediate manual control of the vehicle.” Google states that they take a conservative approach to recognizing disengagements as they are trying to gather “…as much data as possible to enable us to improve our self-driving system.”
Examples of the failure of autonomous technology include broken wires that lead to failure between primary and back-up self-driving systems, sensor reading anomalies related to accelerometers or GPS or anomalies in the monitoring of key functions like steering and braking. For safe operation of the vehicle drivers “…are directed to take control of the vehicle as often as they feel is necessary and for a variety of reasons relating to the comfort of the ride, the safety of the vehicle, or the erratic or unpredictable behavior of other road users.” Afterwards, the safe operation disengagements are replicated in thousands of variations in a simulator to fine tune or fix problems in the self-driving system.
In total the self-driving fleet drove over 424,000 miles on California public roads. The majority of the miles were logged on streets rather than highways because these are typically the most complex driving environments and present the biggest challenge for self-driving cars. A total of 341 disengagements were reported during the period covered by the report with 272 related to failure of the technology and 69 to safe operation of the vehicle. The following chart shows that miles driven per disengagement as it relates to autonomous technology failure has increased during the test period.
The following chart shows the trend of miles driven per disengagement related to to safe operation of the vehicle. Google attributes the lack of a clear trend in the data to the low absolute number of disengagements which can lead to one month skewing the data.
The report provides more detail on the different classifications of disengagement and performance of specific vehicle in the fleet. Learn more: Self-Driving Car Report
Comment: It appears that it is a case of when and not if self-driving cars will become widely available. While LSVs may appear at first to be a safer first step for implementing the technology by virtue of their low speed, the complex environments in which they operate may make them the most challenging. There is also the not so small issue of the cost of self-driving technology in relation to the total cost of the vehicle. As demonstrated by what is occurring in the highway capable automobile market, the most expensive vehicles tend to be the first to include aspects of self-driving technology such as lane detection and auto-braking. It may be the case that LSVs will provide an excellent test bed for the technology but will be the last in line to incorporate the technology once it becomes mainstream.