Methods and models for safety benefit assessment of advanced driver assistance systems in car-to-cyclist conflicts
Doctoral thesis, 2022
Today, there are many components of virtual safety assessment simulations with models or methods that are missing or can be substantially improved. This is particularly true for simulations assessing ADASs that address crashes involving cyclists—a crash type that is not decreasing at the same rate as the overall number of road crashes in Europe. The specific methodological gaps that this work addresses are: a) computational driver models for car-to-cyclist overtaking, b) algorithms for model fitting and efficient calculation of ADAS intervention time, and c) a method for merging data from different data sources into the safety assessment.
Specifically, for a), different driver models for everyday driver behaviour while overtaking cyclists in a naturalistic driving setting were derived and compared. For b), computationally efficient algorithms to fit driver models to data and compute ADAS intervention time were developed for different types of vehicle models. The algorithms can be included in ADAS both for offline use in virtual assessment simulations and online real-time use in in-vehicle ADAS. Lastly, for c), a method was developed that uses Bayesian statistics to combine results from different data sources, e.g., simulations and test-track data, for ADAS safety benefit assessment.
In addition to presenting five peer-reviewed scientific publications, which address these issues, this compilation thesis discusses the use of different data sources; introduces the fundamentals of Bayesian inference, linear programming, and numerical root-finding algorithms; and provides the rationale for methodological choices made, where relevant. Finally, this thesis describes the relationships among the publications and places them into context with existing literature.
This work developed driver models for the virtual simulations and methods for the reliable estimation of the prospective safety benefit, which together have the potential to improve the design and the evaluation of ADAS in general, and ADAS for the car-to-cyclist overtaking scenario in particular.
Traffic safety
overtaking manoeuvres
cyclist
safety benefit
naturalistic data
Author
Jordanka Kovaceva
Chalmers, Mechanics and Maritime Sciences (M2), Vehicle Safety
On the importance of driver models for the development and assessment of active safety: A new collision warning system to make overtaking cyclists safer
Accident Analysis and Prevention,;Vol. 165(2022)
Journal article
Safety benefit assessment of autonomous emergency braking and steering systems for the protection of cyclists and pedestrians based on a combination of computer simulation and real-world test results
Accident Analysis and Prevention,;Vol. 136(2020)
Journal article
A comparison of computational driver models using naturalistic and test-track data from cyclist-overtaking manoeuvres
Transportation Research Part F: Traffic Psychology and Behaviour,;Vol. 75(2020)p. 87-105
Journal article
Drivers overtaking cyclists in the real-world: evidence from a naturalistic driving study
Safety Science,;Vol. 119(2019)p. 199-206
Journal article
Kovaceva, J., Murgovski, N., Kulcsár, B., Wymeersch, H., Bärgman, J. Critical zones for comfortable collision avoidance with a leading vehicle.
The simulations and ADAS addressing car-to-cyclist overtaking situations require computational driver models to make important decisions—so this thesis developed them, by studying drivers’ everyday behaviour while they were overtaking cyclists. It also developed algorithms for model fitting (by making sure that the driver models give the same results as actual data collected from drivers) and for telling the ADAS when to intervene (so it is not so early it is annoying, and not so late that the crash happens anyway). Computing the ADAS intervention time accurately can be achieved by including detailed vehicle models. However, models with high level of complexity require high computation time, so we had to develop highly efficient algorithms for rapid safety benefit assessment. To improve ADAS models overall, a method was created so that data from both simulations and physical tests can be included in the safety assessment. (Different sources provide different information, so putting it all together is not straightforward.) Being able to use information from both sources means we can derive more accurate and robust conclusions regarding the safety benefit of ADAS.
The models and methods developed in this work can help developers get better ADAS onto the market; society will benefit from fewer crashes and cyclists’ injuries.
Proactive Safety for Pedestrians and Cyclists (PROSPECT)
European Commission (EC) (EC/H2020/634149), 2015-05-01 -- 2018-10-31.
eUropean naturalistic Driving and Riding for Infrastructure & Vehicle safety and Environment (UDRIVE)
European Commission (EC) (EC/FP7/314050), 2012-10-01 -- 2016-09-30.
MICA - Modelling Interaction between Cyclists and Automobiles
VINNOVA (2017-05522), 2018-03-09 -- 2019-12-31.
IRIS: Inverse Reinforcement-Learning and Intelligent Swarm Algorithms for Resilient Transportation Networks
Chalmers, 2020-01-01 -- 2021-12-31.
Modellering av Interaktion mellan Cyklister och Fordon 2- MICA2
VINNOVA (d-nr2019-03082), 2019-11-01 -- 2022-12-31.
Areas of Advance
Transport
Subject Categories
Vehicle Engineering
ISBN
978-91-7905-696-4
Doktorsavhandlingar vid Chalmers tekniska högskola. Ny serie: 5162
Publisher
Chalmers
Room Omega, Building Jupiter, Hörselgången 5, Chalmers Campus Lindholmen
Opponent: Associate Professor Shan Bao, Department of Industrial and Manufacturing Systems Engineering, University of Michigan-Dearborn, Michigan