LADRI: LeArning-Based Dynamic Risk Indicator in Automated Driving System

EasyChair Preprint 11096

Abstract

The framework presented herein introduces a methodology for Dynamic Risk Assessment (DRA) in Automated Driving Systems (ADS), utilizing the prowess of Artificial Neural Networks (ANNs). With the escalating progression towards intelligent transportation and autonomous driving systems, the necessity to instate robust safety measures has never been more critical. Traditional risk assessment techniques, optimized for human-operated vehicles, prove insufficient to cater to the dynamically changing environments ADS operate in. This underlines the need for real-time DRA, which equips the ADS to comprehend its immediate risk landscape and adapt its decision-making accordingly.

The proposed solution leverages ANNs, a prominent branch of deep learning, to meticulously discern and categorize risk levels of severity and controllability using On-board Sensor (OBS) data. ANNs have shown commendable potential in managing an array of challenges posed by ADS by efficiently sifting through voluminous and intricate data to recognize underlying patterns. This learning-based methodology, by scrutinizing OBS data, enables a precise determination of the current risk quotient, thereby heightening the situational awareness of the ADS.

This enhanced awareness augments the system's understanding of its operating context and surroundings, significantly improving the safety of both the passengers within and the external traffic entities. The proposed solution also addresses the shortcomings of traditional risk assessment techniques by harnessing the capabilities of ANNs, thereby empowering ADS to accurately gauge the risk factors that potentially lead to accidents.

Keyphrases: Artificial Neural Network, Automated Driving System, dynamic risk assessment

@booklet{EasyChair:11096,
  author    = {Anil Ranjitbhai Patel and Peter Liggesmeyer},
  title     = {LADRI: LeArning-Based Dynamic Risk Indicator in Automated Driving System},
  howpublished = {EasyChair Preprint 11096},
  year      = {EasyChair, 2023}}