City Research Online

Abstract

This code, written for MATLAB (compatible with versions from 2019 onwards), has been used in the preparation of a journal article "Why Black-Box Bayesian Reliability Assessment of Autonomous Systems is Problematic and what can be done about it?". The code allows one to conduct a Bayesian safety assessment of an autonomous vehicle used on the public roads with different operating conditions (commonly referred to as an Operational Design Domain). The assessment can be done using a black - box model (ignoring the existence of different operating conditions) and a white - box model which takes account of the operating conditions for which the observations are collected.
The code includes the computation procedures (including univariate probability distributions to capture the prior beliefs about the probability of failure per mile in each of the defined conditions and a Dirichlet distribution which captures the assessor uncertainty about the likelihood of different operating conditions). The code includes also examples of comparison between the predictions made with a black-box and a white-box model for different observations.

Publication Type:	Software
Publisher Keywords:	Autonomous vehicle, Bayesian inference, “driving to safety”
Subjects:	Q Science > QA Mathematics > QA75 Electronic computers. Computer science
Departments:	School of Science & Technology School of Science & Technology > Computer Science
SWORD Depositor:	Symplectic Administrator

Export

Downloads