RATIONALE & INTRODUCTION:

Automated driving system (ADS) developers need a way to describe safe and competent driving for automated driving system-dedicated vehicles (ADS-DVs) in a way that is relatable to how stakeholders interpret safe driving today. Metrics informed by competent and safe human behavior could improve understanding and confidence in ADS-DVs. One way to make ADS safety performance relatable to stakeholders is to adopt an intuitive comparison to behaviors displayed on the road by human drivers. 

This best practice outlines a process for leveraging human driving data to establish safety performance targets for ADS-DV behaviors. The targets within the specific use-case exemplified in this best practice are based on naturalistic driving data from manually driven vehicles in the hope of aiding understanding from a broad audience of stakeholders.
The safety performance of some ADS behaviors can be measured and compared to naturalistic driving studies (NDS) data from human drivers to help characterize the socially acceptable balance between safety, lawful driving, efficiency, and comfort. For instance, developers can utilize human-driver data to determine an appropriate minimum passing distance when vulnerable road users (VRUs) are present. This process enables them to enhance the safety performance of ADS fleets by aligning it with human-relative benchmarks and considerations.

The recommended set of safety performance metrics from previous best practice AVSC00006202103 and the behavioral competencies from AVSC00008202111 provide a starting point for assessing the ADS behavior in a specific operational design domain (ODD). This best practice builds upon those previous efforts by providing ADS developers with a third component for safety assurance by identifying:
• The specific behavioral competency of interest for evaluation.
• The applicable safety performance metrics.
• The data that can be leveraged to define safety performance reference values.

SCOPE 

Human driving data provides one possible source for establishment of reference values (i.e., performance threshold) for ADS-DVs. This best practice describes a framework to establish human driving data reference values for ADS safety performance metrics. This framework provides a way to assess different behaviors and is broadly generalizable. Key characteristics of datasets that can be used to establish these reference values for ADS are described in an objective, repeatable, and explainable way.

This best practice offers guidance on ensuring data quality and conducting appropriate analyses, including sample size determination, error analysis and interpretation, variance, standard deviation, segmentation, and normalization. The guidelines established in this best practice specifically concentrate on interactions among road users, such as vehicles and pedestrians. Other factors (e.g., presence of objects like debris or construction equipment) and other aspects of safety behavior that do not involve the interaction between two road users are beyond the scope of this best practice.