Damerow, Florian (2018)
Situation-based Risk Evaluation and Behavior Planning.
Technische Universität Darmstadt
Ph.D. Thesis, Primary publication
|
Text
Diss.pdf - Accepted Version Copyright Information: CC BY-ND 4.0 International - Creative Commons, Attribution NoDerivs. Download (8MB) | Preview |
Item Type: | Ph.D. Thesis | ||||
---|---|---|---|---|---|
Type of entry: | Primary publication | ||||
Title: | Situation-based Risk Evaluation and Behavior Planning | ||||
Language: | English | ||||
Referees: | Adamy, Prof. Dr. Jürgen ; Sendhoff, Prof. Dr. Bernhard ; Hochberger, Prof. Dr. Christian ; Griepentrog, Prof. Dr. Gerd | ||||
Date: | 2018 | ||||
Place of Publication: | Darmstadt | ||||
Date of oral examination: | 14 August 2017 | ||||
Abstract: | The presented dissertation addresses the problem of risk evaluation and behavior planning for future intelligent Advanced Driver Assistance Systems (ADAS). For this purpose, a novel framework for situation-based risk evaluation and behavior planning, targeting highly automated driving, is presented. After properly sensing the current scene, including the current road topology and other traffic participants, the proposed framework first estimates and predicts the future behavior of all involved entities comprising a situation classification and trajectory prediction step. This is then followed by the generation of the own future behavior in a behavior planning step which is based on an evaluation of possible ego behavior alternatives in terms of risk and utility considerations. The future behavior is planned in a way to find a tradeoff between the expected future risk and utility. Inner-city traffic scenarios in particular are usually complex and of high uncertainty, considering measurements as well as behavioral decisions. To reduce the complexity, similar behavior alternatives are clustered and represented by prototypical behavior patterns using so-called situations. A novel situation classification approach is proposed to estimate how good a situation matches with the actual behaviors. This approach is based on a comparison of the prototypically predicted trajectories of the considered situations with the actual measured trajectories. For this purpose a novel measure for spatio-temporal trajectory similarity, based on the evaluation of longitudinal and lateral spatio-temporal distance, is derived. The situation classification system is used to detect incorrect and critical traffic behaviors, especially in scenarios with a disregard of right-of-way. Evaluating the system using real-world crash cases reveals that it is able to warn the driver reliably of an upcoming crash, with sufficient time to initiate a suitable evasive behavior. For the prediction of situation-dependent prototypical scene evolution patterns, the interaction-aware Foresighted Driver Model (FDM) is applied in a forward simulation of a sensed scene under different situation-dependent behavioral assumptions. The proposed FDM is a novel, time continuous driver model for the simulation and prediction of freeway and urban traffic. Based on the general risk evaluation and behavior planning framework developed in this thesis, the driver model equations are introduced from the assumption that a driver tries to balance predictive risk (e.g. due to possible collisions along its route) with utility (e.g. the time required to travel, smoothness of ride, etc.). For this purpose, a computationally inexpensive, approximate risk model targeting only risk maxima and a gradient descent-based behavior generation is applied. It is shown, how such a model can be used to simulate and predict driving behavior with a similar performance compared to full behavior planning models. The FDM is applicable to a wide range of different scenarios, e.g. intersection or highway-accessing scenarios, with the consideration of an arbitrary number of traffic entities. Thus, the FDM generalizes and reaches beyond state-of-the-art driver models. Complex traffic situations require the estimation of future behavior alternatives in terms of predictive risks. Risk assessment has to be driven from the knowledge that the acting scene entity requires to evaluate the own future behavior. Based on the predicted future dynamics of traffic scene entities, an approach is presented, where a continuous, probabilistic model for future risk is used to build so-called predictive risk maps. These maps indicate how risky a certain ego behavior will be at different future times, so that they can be used to directly plan the best possible future behavior. The behavior in complex scenarios differs strongly, depending on the actually occurring situation. However, sensory measurements of the ego- and other involved entities' states as well as the prediction of possible future states are generally of high uncertainty. As a consequence, the current driving situation can only be approximated. Additionally, a situation can change very quickly, e.g. if a traffic participant suddenly changes its behavior. In this thesis an approach is proposed, how to plan a safe, but still efficient future behavior under consideration of multiple possible situations with different occurrence probabilities. In several traffic scenarios comprising simulated as well as recorded real-world data, it is shown that the approach generates an efficient behavior for situations which are likely to occur, while generating a plan B to safely deal with improbable but risky situations. |
||||
Alternative Abstract: |
|
||||
URN: | urn:nbn:de:tuda-tuprints-67903 | ||||
Classification DDC: | 600 Technology, medicine, applied sciences > 620 Engineering and machine engineering | ||||
Divisions: | 18 Department of Electrical Engineering and Information Technology 18 Department of Electrical Engineering and Information Technology > Institut für Automatisierungstechnik und Mechatronik > Control Methods and Robotics (from 01.08.2022 renamed Control Methods and Intelligent Systems) |
||||
Date Deposited: | 16 Feb 2018 14:44 | ||||
Last Modified: | 09 Jul 2020 01:51 | ||||
URI: | https://tuprints.ulb.tu-darmstadt.de/id/eprint/6790 | ||||
PPN: | 426510151 | ||||
Export: |
View Item |