AV Regulations: What’s the Hold-Up?
June 4, 2021—Autonomous vehicle (AV) stakeholders all seem to agree on the need for operational safety assessment (OSA) metrics to safely deploy AVs on public roads and to understand impact, but there is not yet consensus on what OSA metrics should be used.
That’s according to a May webinar from the University of Michigan’s Connected and Automated Vehicles and Infrastructure Global Symposium.
Among the factors to consider were:
- Data access
- Operational design domain (ODD) coverage, which would define where AVs can function based on current capabilities
- Kinematic components of the dynamic driving task (meaning all driver tasks are carried out by the driver)
- The need to minimize false negatives and false positives
- Validation, including with real-world data
- Comfort metrics
At this time, U.S. federal regulations on OSA metrics do not exist, although the concept was recently mentioned in Advanced Notice of Proposed Rulemaking.
Several states have implemented various OSA metrics, such as California’s disengagement metric.
In April 2020, the United Nations Economic Commission for Europe (UNECE) released “Uniform provisions concerning the approval of vehicles with regard to automated lane keeping systems,” which was the world’s first regulation of a driving automation feature. More than 65 countries adopted those regulations.
The regulation has operational safety requirements for simple scenarios such as car following, cut-in, cut-out revealing stationary object, led vehicle decelerating, and limits the speed of the feature to 60 kilometers per hour. In addition, the regulations require automated driving system (ADS) operation data storage.
That approach, however, has limitations. The webinar participants noted that those limitations included that it:
- Does not consider reaction time of the subject vehicle.
- Does not consider the velocity of an adversarial vehicle.
- Does not consider acceleration or acceleration/deceleration capability of either subject or adversarial vehicle.
- Basic formulation applicable only in longitudinal-based scenarios.
Regardless, the participants noted that it was a positive first step in regulation of AVs, including specifying OSA metrics. However, the lack of comprehensive regulations means that false negatives in certain scenarios are likely.
That’s why the SAE On-Road Automated Driving (ORAD) Committee was created to address the challenges associated with ADS development. That committee has been responsible for SAE J3016, the consensus taxonomy for driving automation levels. Within that committee, the Verification and Validation Task Force is developing J3237 on OSA metrics.
That project is still a work in progress, but as the year progresses, it will become critical to understand the proposed metrics, as it will likely lead the way for AV development.