Towards a Standardized Framework for Reporting and Analyzing Autonomous Vehicle Disengagements

EasyChair Preprint 11656

Abstract

This paper explores the critical aspect of Autonomous Vehicle Disengagements (AVDs) in the development and testing of autonomous vehicles (AVs). Divided into three parts, the paper begins by emphasizing the significance of standardized reporting for AVDs, which are instances when control is handed back to human drivers during autonomous operation. The reporting framework encompasses criteria definition, data collection, classification, and continuous improvement loops. It delves into industry-wide and manufacturer-specific trends, highlighting challenges and threats posed by AVDs. This paper provides a thorough analysis of technical, operational, regulatory, and ethical challenges associated with AVDs. It identifies potential safety threats and emphasizes the importance of transparent reporting and collaboration between industry stakeholders and regulatory bodies. The trends in AVD reports are examined globally, emphasizing the variability across manufacturers, geographic locations, and testing environments. Furthermore, the paper suggests implications and future directions for the AV industry, including root cause analyses, continuous refinement of perception algorithms, collaboration with regulatory agencies, and enhancing ethical decision-making frameworks. It addresses the integration of AVs into existing transportation systems, proposing measures for infrastructure adaptation, public awareness, interoperability, and data privacy. The conclusion highlights the need for standardized safety metrics, further research on driver reactions, and the exploration of human-machine interfaces for a safer and more efficient deployment of autonomous vehicles.

Keyphrases: Autonomous, Disengagements, vehicle

@booklet{EasyChair:11656,
  author    = {Jane Smith and Patrick Oric},
  title     = {Towards a Standardized Framework for Reporting and Analyzing Autonomous Vehicle Disengagements},
  howpublished = {EasyChair Preprint 11656},
  year      = {EasyChair, 2024}}