On June 18, Chris van der Ploeg will publically defend his Ph.D. dissertation. The public defense starts at 16:00, which will commence with a brief presentation.
Download dissertation
The dissertation is available for download through this link
The defense will take place in the Atlast building, room 0.710.
Address: Het Eeuwsel 53, 5612 AZ Eindhoven
Summary
In the pursuit of zero traffic casualties and higher levels of vehicle automation, new challenges emerge as more tasks are automated. Despite significant advances in automated vehicle development over the last few decades, they are still susceptible to conditions that were traditionally the responsibility of human drivers to detect and react to. These conditions can appear in different parts of the vehicle such as hardware components, software running on the vehicle, or signals transmitted or received from the vehicle. They can also arise from the vehicle's behavior or its interpretation of the environment. This necessitates the creation of approaches for the vehicle to understand its own condition and operation within a given setting, encompassing its performance, operational boundaries, and the consequences of its actions. Furthermore, it should possess the capability to sense, comprehend, and predict elements and occurrences in its immediate environment.
The aim of this thesis is to improve the advancement of automated driving technology by addressing some of the current obstacles in these fields of research.
The first part of the thesis concentrates on detecting and estimating faults in dynamical systems. The developed methods are applied to the steering system of an automated vehicle for which, from a practical point of view, it is crucial to be able to detect and estimate any faults that may arise to allow safe mitigation and therefore safe operation.
A methodology for fault detection and estimation is proposed that enables the simultaneous detection and estimation of additive and multiplicative faults in linear time-invariant dynamical systems. The main difficulty addressed in this study is the dynamic inseparability of these faults, which means that they influence the system through the same input/output relationship. First, a residual generator is proposed that detects and estimates the combined effect of additive and multiplicative faults. Subsequently, the individual effect of these faults is estimated using a nonlinear regression operator. The proposed filter architecture is accompanied with performance bounds, allowing the user to be aware of the error margin of the fault estimates and the possible adjustments that can be made to the filter or system to improve the estimation performance. In simulations, the results show that the approach allows for the detection and estimation of the desired faults.
In an experimental setting, real-world effects may occur that are not taken into account in a simulated environment. To address this problem, two research directions are explored to incorporate these effects into the detection and estimation of faults for automated vehicles. The first direction focuses on incorporating measurable changes in parameters within the system. For example, in the case of an automated vehicle, a gradually changing longitudinal velocity can transform the description of the system from a linear time-invariant system to a linear parameter-varying system. This change in system behavior can result in wrong detections and estimates of the fault when using a linear time-invariant approach. To address this issue, a methodology is proposed to design a residual generator for the linear parameter-varying filter that is scheduled based on the measurable parameter.
The second approach aims to reduce the impact of unmeasurable variations in system parameters, as well as the effects of working with digital systems, such as measurement noise and delays. An offline simulator allows us to numerically characterize the mismatch between an assumed linear model and a range of alternative linear models that exhibit different levels of structured uncertainty. Moreover, we show how the performance bounds of the estimator, valid in the absence of uncertainty, can be used to determine appropriate countermeasures for measurement noise. The efficacy of the proposed approach is demonstrated in simulations and in-vehicle experiments.
To increase autonomy in more difficult and complex driving cases, the second part of the thesis focuses on safe motion planning for an automated vehicle. This includes staying within the designated lane and taking into account observed and potentially occluded road users. These occlusions can arise from, for example, sensor faults, adverse weather conditions, or physical obstructions. A motion planning strategy is presented that prioritizes safety considering various factors such as observed objects, predictions, road information, and the reachable set of potentially hidden objects. These factors are then mapped to repulsive fields that could indicate the risk of harm in the event of a collision or the potential dangerous consequences of certain actions. Incorporating these risk fields with driving objectives, such as navigating towards a particular point of interest, enables a model-predictive planning strategy that considers a multi-objective trade-off between these criteria. In simulation, the proposed approach shows anticipative behavior in the presence of obstructions, preventing collisions with vulnerable road users and vehicles that could not have been anticipated without the reachable set prediction of possibly hidden objects.
The contributions introduced in this thesis enhance the level of self-awareness and situational awareness for automated vehicles. This brings us closer to the safe commercial deployment of highly automated vehicles, and a more responsible introduction of these on public roads.