Safety performance of the waymo rider-only automated driving system at one million miles

This paper examines the safety performance of the Waymo Driver™, Waymo’s Automated Driving System (ADS). It analyzes one million miles of driving on public roads in parts of California and Arizona with no human behind the wheel – what we call rider-only (RO) operations. There were no reported injuries, and only two collisions that were comparable to the National Highway Traffic Safety Administration’s Crash Investigation Sampling System (“CISS”), a nationally representative collision database of collisions that were police reported and in which at least one vehicle was towed. There were an additional 18 minor-contact events that were too minor to meet the tow-away and police-report criteria for CISS, where nine of these 20 contact events had no damage. Neither the CISS-comparable- nor the minor-contact events were intersection related or involved VRUs (Vulnerable Road Users), and every vehicle-to-vehicle event involved one or more road rule violations and/or dangerous behaviors on the part of the other vehicle’s operator. 55% of all contact events occurred with a stationary Waymo vehicle, and 40% were parking related, with some overlaps between the two categories. These results, in combination with previous reported data and past studies performed by Waymo on fatalities reconstruction and on collision-avoidance testing, support the assertion that the Waymo Driver is successful at reducing injuries and fatalities today. To establish a valid comparative assessment of the Waymo Driver, it is most efficient to apply the same requirements to ADS and Human comparison, only counting events that are likely included in human-reported crash data benchmark, for example, only CISS-compatible collisions. Confidence in conclusions about the safety of the Waymo Driver will increase continuously as both credibility of available data and the validity of predictions improves over time.