IIHS What humanlike errors do autonomous vehicles need to avoid to maximize safety?

A discussion paper, based on an extensive analysis of on-road vehicle incidents of the requirements for an Autonomous Vehicle to perform better than a human vehicle operator.

This meta-study summarised the causes of human error based on the Endsley model with an additional parameter of Incapacitation. The distribution of accidents attributable to each type is summarised as part of informing functional requirements development for any AV (Autonomous Vehicle) as:

Sensing/Perceiving with 24% (plus an additional 10% for human incapacitation)
Predicting at 17%
Planning/Deciding at 39% impacted by speed (24%) or illegal maneuvers (15%), and
Execution/Performance at 23%

The analysis is notable in that it does not base its conclusions on single, dominant causes – but includes input from all causes in developing the required performance of any AV system.