From Neuron Coverage to Steering Angle: Testing Autonomous Vehicles Effectively

Published

August 2, 2021

Author(s)

Jack Toohey, M S Raunak, Dave Binkley

Abstract

A Deep Neural Network (DNN) based system, such as the one used for autonomous vehicle operations, is a "black box" of complex interactions resulting in a classification or prediction. An important question for any such system is how to increase the reliability of, and consequently the trust in, the underlying model. To this end, researchers have largely resorted to adapting existing testing techniques. For example, similar to statement or branch coverage in traditional software testing, neuron coverage has been hypothesized as an effective metric for assessing a test suite's strength toward uncovering failures and anomalies in the DNN. We investigate the use of realistic transformations to create new images for testing a trained autonomous vehicle DNN, and its impact on neuron coverage as well as the model output.

Citation

Special Issue on Safety, Security, and Reliability of Autonomous Vehicle Software

Volume

54

Issue

8

Pub Type

Journals

Download Paper

https://doi.org/10.1109/MC.2021.3079921

Local Download

Keywords

autonomous vehicles, neural network, metamorphic testing

Citation

Toohey, J. , Raunak, M. and Binkley, D. (2021), From Neuron Coverage to Steering Angle: Testing Autonomous Vehicles Effectively, Special Issue on Safety, Security, and Reliability of Autonomous Vehicle Software, [online], https://doi.org/10.1109/MC.2021.3079921, https://tsapps.nist.gov/publication/get_pdf.cfm?pub_id=932505 (Accessed December 1, 2024)