RSS Model Calibration and Evaluation for AV Driving Safety based on Naturalistic Driving Data

Yiwen Huang, Maria Soledad Elli, Jack Weast, Yingyan Lou, Shi Lu, Yan Chen

Research output: Contribution to journalConference articlepeer-review

Abstract

The definition and evaluation of driving safety for automated vehicles (AVs) is a key element to enable safe and scalable AV applications. Although different distance-based and/or time-based safety metrics have been proposed, a unified standard has not yet been recognized as a baseline to assess the safety of AV behavior. Moreover, the relationship between existing AV safety metrics (models) and naturalistic driving data, which indicate driving safety and comfort defined by human drivers, is not well explored. Utilizing the responsibility-sensitive safety (RSS) model, a new methodology is proposed to calibrate the RSS model based on naturalistic driving data. Without significantly relying on (large) safety-critical or collision data, the proposed method defines an optimization framework to calibrate the RSS model parameters and describes AV driving safety through both safe and safety-critical data in a cross-checking manner. Evaluation of the calibrated RSS model is discussed based on naturalistic driving data in Los Angeles, USA.

Original languageEnglish (US)
Pages (from-to)430-436
Number of pages7
JournalIFAC-PapersOnLine
Volume54
Issue number20
DOIs
StatePublished - Nov 1 2021
Externally publishedYes
Event2021 Modeling, Estimation and Control Conference, MECC 2021 - Austin, United States
Duration: Oct 24 2021Oct 27 2021

Keywords

  • Automated vehicles
  • Model calibration
  • Naturalistic driving
  • Optimization
  • Responsibility sensitive safety

ASJC Scopus subject areas

  • Control and Systems Engineering

Fingerprint

Dive into the research topics of 'RSS Model Calibration and Evaluation for AV Driving Safety based on Naturalistic Driving Data'. Together they form a unique fingerprint.

Cite this