Müller, Fabian (2023)
Severity Estimation for Risk-based Motion Planning.
Technische Universität Darmstadt
doi: 10.26083/tuprints-00023325
Ph.D. Thesis, Primary publication, Publisher's Version
Text
Dissertation_Fabian_Mueller_final.pdf Copyright Information: CC BY-SA 4.0 International - Creative Commons, Attribution ShareAlike. Download (3MB) |
Item Type: | Ph.D. Thesis | ||||
---|---|---|---|---|---|
Type of entry: | Primary publication | ||||
Title: | Severity Estimation for Risk-based Motion Planning | ||||
Language: | English | ||||
Referees: | Adamy, Prof. Dr. Jürgen ; Sendhoff, Prof. Dr. Bernhard | ||||
Date: | 2023 | ||||
Place of Publication: | Darmstadt | ||||
Collation: | XXV, 235 Seiten | ||||
Date of oral examination: | 16 December 2022 | ||||
DOI: | 10.26083/tuprints-00023325 | ||||
Abstract: | The goal of autonomous driving is to increase safety, benefit and comfort for all road users. Above all, the area of risk perception is of central importance in preventing critical situations or averting possible harm. Factors such as measurement uncertainties in environment perception, uncertainties about the future behavior of road users change the probability of occurrence of critical events, such as a collision between road users, and thus the risk of planned driving maneuvers. In this work, based on an environmental representation and given motion models, a risk assessment for a discrete-time event prediction with focus on collisions between two vehicles is presented as part of a cost-based planner, which additionally adds utility and comfort along a planned trajectory. A risk assessment includes not only the often modeled probability of occurring critical events, but also their damages such as injuries to occupants or loss of value of the vehicle. Classical metrics consider either accident probability or severity. In this work, both components of risk are considered together, whose impact on driving behavior becomes visible in medium-critical scenarios such as overtaking or passing in narrow scenarios. The modeling of a collision event considers two polygonal-shaped objects from the bird's eye view - preferably rectangles - of different sizes. States like positions and velocities of the objects are subject to uncertainties which are approximated by a Gaussian distribution. In order to detect all collisions even with highly dynamic objects, the detection is quasi time-continuous. This means that in addition to checking at discrete sampling time points, collision constellations located between two consecutive time points are also considered. Furthermore, analytical methods for the determination of the collision probability and of all state distributions are presented, which either represent states with collided traffic participants or include only collision-free trajectories. Compared to a classical Monte Carlo simulation with 1000 samples, the computation time is significantly reduced while maintaining the same accuracy. To further increase the accuracy, the state distribution of the non-collided trajectories is represented using multidimensional Gaussian Mixture Models in state space, whose number of combined unimodal components depends on possible avoidance scenarios. A modeling approach is presented for the severity of a collision that considers the injuries of subsequent collision events in addition to the injury severity of the initial contact. Overall, the accident severity model includes the following components: a posteriori states after the collision by applying the momentum conservation equations, the control capability of the vehicles after the collision, the injury of the vehicle occupants during the collision, and an ethical trade-off between the injury probabilities of all involved occupants. To evaluate the presented model, it is subsequently compared with three less detailed models, which are strongly based on the literature. In the simulations of critical driving scenarios, it is shown, among other things, that when accident severity models are used, a speed-adaptive transition range is established between simple following scenarios and narrow overtaking maneuvers due to the interaction between severity and collision probability evaluations. Here, a speed difference is formed that creates a minimum risk during the overtaking maneuver and is primarily dependent on the lateral distance. In addition, the application of the momentum conservation equation with its masses and the injury modeling of all occupants leads to the protection of the weaker collision partner between dissimilar vehicles, such as a car versus a truck, which is expressed by a significant reduction in overtaking speeds. |
||||
Alternative Abstract: |
|
||||
Status: | Publisher's Version | ||||
URN: | urn:nbn:de:tuda-tuprints-233259 | ||||
Classification DDC: | 600 Technology, medicine, applied sciences > 620 Engineering and machine engineering | ||||
Divisions: | 18 Department of Electrical Engineering and Information Technology > Institut für Automatisierungstechnik und Mechatronik > Control Methods and Intelligent Systems | ||||
Date Deposited: | 14 Mar 2023 13:01 | ||||
Last Modified: | 14 Mar 2023 14:03 | ||||
URI: | https://tuprints.ulb.tu-darmstadt.de/id/eprint/23325 | ||||
PPN: | 505929627 | ||||
Export: |
View Item |