his thesis introduces a Model Predictive Fuzzy Control (MPFC) framework for mission-planning in multi-agent systems operating in dynamic, uncertain environments. MPFC integrates predictive decision-making from Model Predictive Control (MPC) with the adaptability of Fuzzy Logic Control (FLC) to handle complex, time-sensitive tasks. A generic mathematical formulation is developed and applied to a multi-agent Search and Rescue case study with fire and wind dynamics. Comparative performance analysis against MPC and Pre-Tuned FLC controllers, along with sensitivity analyses and design exploration, demonstrate that MPFC can outperform existing approaches and maintain robust performance under varying conditions. Although this technique is promising, several limitations and challenges are identified, suggesting avenues for future research to refine and enhance its applicability.

Marine pollution is a critical issue impacting the global community, with underwater waste a particularly daunting challenge. While autonomous detection and collection of underwater waste is highly desirable, these are extremely difficult tasks. This difficulty arises from the intrinsic complexities of the aquatic environment, including variable lighting conditions, reduced visibility, and the complex nature of water currents. This paper focuses on novel approaches for autonomous detection of underwater waste and proposes to incorporate domain knowledge to refine deep-learning-based underwater object detection techniques. More specifically, the domain knowledge is represented with models in the state space form that describe the motion of the target objects to assist the classification of objects. Moreover, optical flow is combined with a k-means clustering algorithm to extract the trajectory of the target objects from videos. These trajectories are subsequently fed into a neural network that is trained using knowledge distillation, enhanced with domain knowledge. For our experiments, a simulator is devised to facilitate the creation of a dataset for developing and testing the proposed architecture. The results of the experiments demonstrate that including domain knowledge within the object detection approach with neural networks provides numerous substantial advantages, including enhanced robustness against noisy and poorly labelled data, facilitation of semi-supervised learning, and consistent superiority in accuracy over the baseline scenario. Additionally, combining the domain knowledge with a neural network significantly increases the computational speed of object detection compared to using the standalone domain knowledge module.

Search and Rescue (SaR) missions present challenges due to the complexity of the disaster scenarios. Most life losses and injuries occur in developing countries. Robotics has become indispensable for rapidly locating disaster victims. Combining flying and ground robots more effectively serves this purpose due to their complementary features. To this end, a cost-effective framework to perform conventional SaR tasks is presented. The method leverages You Only Look Once and video streams by an Unmanned Ground Vehicle (UGV) and an Unmanned Aerial Vehicle (UAV). In exploiting pose estimation to perform human depth estimation, the susceptibility of the algorithm to variations in poses was unveiled. In tracking object trajectories, the collaboration is advantageous in wide-area cluttered trajectories as opposed to narrow-area unobstructed trajectories. In mapping terrain elevation, errors drop significantly with the assistance of the UAV. Moving forward, devising adaptable strategies tailored to diverse SaR scenarios will be pivotal.

Tiny object detection (TOD) in satellite imagery is critical for applications including pipeline monitoring, where the detection of tiny objects, such as excavators near the pipeline networks, can prevent potential incidents. However, TOD faces challenges due to the limited pixel representation of objects, complications with Intersection over Union-based label assignment, and semantic confusion between visually similar classes. This paper introduces OE-Net, a modular extension of Oriented R-CNN,
specifically designed for TOD in satellite images. OE-Net integrates three novel components: a Dilated Fusion module to enhance the feature extraction, a Balanced Ranking Assigner to improve the anchor matching, and Vector Mapping for more precise classification of visually similar objects. The modular design of OE-Net allows its novel components to be easily integrated into other detection models. Tested on the TinyDOTA dataset, OE-Net sets a new benchmark in TOD performance, achieving a 6.3
percentage points improvement in mean Average Precision (mAP) over the Oriented R-CNN baseline. On a novel excavator detection dataset, named ExcaSat and developed for pipeline monitoring, OE-Net outperforms the Oriented R-CNN baseline by 11.1 percentage points in mAP. Furthermore, OE-Net surpasses state-of-the-art methods on both TinyDOTA and ExcaSat, establishing a new standard in computational efficiency and detection precision for satellite-based monitoring tasks. The code is available at: https://github.com/OE-jimvanoosten/mmrotate-OE.git.

The impact of disasters on the affected population is catastrophic. Proper disaster management practices are needed to reduce their societal damage. This includes Search and Rescue (SaR) missions, which pertain measures to find potentially trapped victims. This task can be alleviated by the support of a fleet of SaR robots. This fleet is deployed to locate trapped victims and to report back their position to the emergency responders and potentially to follow them to track of their potential movements. Locating trapped victims is challenging, as the number of these people, their initial position and potential displacements may be unknown. A behavioral model of the victims can give insight on how they take decisions and act during an evacuation situation. Several techniques are used in the state of the art. This study proposes an evacuation model that integrates game theory into the belief-desire-intention framework. The model is validated with a benchmark from the state of the art. It is found that the model is able to produce realistic evacuation times. This depends on the distribution of the game theoretic strategies in the population. SaR robots are then added to the validated model. It is found that their presence reduces the evacuation time, depending on several parameters influencing trust of the victims in the robots. Thus, this research contributes to the field of research SaR operations by providing insight into the behaviour of the trapped victims in the presence of rescue robots.

A key challenge for SaR robotics is to avoid dynamic obstacles in cluttered environments, with limited and noisy information. In this research, a controller for SaR robots is developed by coupling a local heuristic motion planner with a model predictive control (MPC) based trajectory tracker. Constraint tightening and tube-based control are used to make the MPC robust to model mismatch and additive measurement noise, while the motion planner is integrated with the MPC. The motion planner periodically supplies a reference trajectory to the trajectory tracker, but the MPC can request additional updates in case of a noticeable mismatch between the predicted and measured environment, based on a user-defined threshold. A case study is designed in MATLAB where a single robot needs to reach a goal through a cluttered environment with dynamic obstacles. Results from the case study show that the MPC method outperforms two state-of-the-art control approaches that are based on the rapidly-exploring random tree (RRT) and artificial potential function (APF) methods. In particular, the heuristic and MPC coupled controller showed a higher success rate in reaching the goal without collisions, and displayed a lower path length in cases with both low and high computational budget.

An aging population puts a pressure on health-care workers working with dementia patients globally. A potential solution is to provide care with Socially Assistive Robots (SARs), i.e. robots who help people through social interaction. However, for effective care these SARs must be able to personalize their behavior to individual patients and adapt this behavior to changes in this patient’s references. This paper presents the decision making process of a SAR that enables the SAR to personalize its behavior to the personality of the patients and adapt this behavior to their current state-of-mind. The system consists of a Fuzzy Logic Control personalization module, which personalizes the SAR’s behavior and a Reinforcement Learning based decision making module together with a Fuzzy Logic Control reward module for the adaption of the personalized behavior. The personalization of the SAR’s behavior is assessed by comparing the output of the system with answers from a survey. The average scatter index over all different behavioral parameters of the SAR is 20.4% The adaptation of the behavior is assessed with computer-based simulations, where an overall accuracy of 81.8% is achieved. A third experiment is carried out to assess the effect of adding the Fuzzy Logic Control personalization module to the system. This experiment shows that adding the personalization module to the decision making system of the SAR decreases the time for the learning process to converge with 13.3%. Although the first assessments of the system look promising, more extensive experiments should be held in later stages of the research. A crucial experiment that must be held in future research is performing real-life interactions between dementia patients and the SAR, in such an experiment the functionality of the system really can be assessed.

The application of autonomous robots in search-and-rescue (SAR) missions forms a challenging field of research. Cooperative search behaviour can greatly increase the efficiency with which a multi-agent system creates situational awareness of and finds victims within an unknown environment. In this research we develop an autonomous mission planning approach that exploits non-homogeneous characteristics of the robots to increase the overall search performance. Furthermore, the proposed approaches incorporate a hierarchical, cooperative control architecture: At the lower level of control, every SAR agent is locally controlled by a heuristic approach that uses fuzzy-logic control (FLC) and A* search to guide its individual actions. At the higher level of control, a model-predictive-control-based (MPC-based) system coordinates the actions of all SAR agents when two or more agents intend on searching the same area at once. When simulated in a virtual SAR environment, we show that the area coverage performance of the cooperative search approach is comparable to a purely heuristic approach designed for area coverage, while simultaneously outperforming this and another optimisation-based approach in victim detection efficiency. This shows that the hierarchical combination of the heuristic local controller and the optimisation-based supervisory controller is able to outperform either one approach individually. Furthermore, it is demonstrated that the cooperative search approach is able to efficiently resolve particular SAR-related search conflicts where a non-cooperative structure fails.

Socially Assistive Robots (SARs) i.e., robots that assist humans through social interactions with them, have shown potential to improve the quality of life of their users. Nonetheless, the state-of-the-art SARs face challenges that prevent them from or limit them in assisting humans in the real world. Although current SARs display cues that simulate social behaviours, the interactions that they generate are often not realistic and fail to engage their users for long periods of time. Most of the challenges currently faced by SARs are a consequence of the lack of social awareness and understanding of the cognitive procedures of individual users. In human interactions, people develop cognitive models of each other to achieve social goals and to respond to the needs of their peers according to their emotional states. Therefore, the present research project proposes to develop a mathematical model of human cognition, based on Theory of Mind (ToM), and to implement it for SARs. To do so, the cognitive processes behind human behaviour in social interactions are carefully investigated and formalised, and a model of human cognition is presented. Two formulations of the model are proposed and compared. The model is implemented using Fuzzy Cognitive Map (FCM), and is then qualitatively analysed and quantitatively assessed. It is concluded that, to accurately estimate the cognitive states of humans, the model must comprise user-specific variables that describe agents throughout interactions, the linkages of the FCM must be represented as functions, and the model must be personalised to every user.

With the number of diagnosed cases rising every year, Autism Spectrum Disorder (ASD) is in need of novel therapeutic approaches to counteract the social and motor impairments inflicted by it. In this research, one of such therapeutic approaches, movement therapy (in particular Dance Movement Therapy (DMT)), is combined with the concept of Socially Assistive Robots (SARs) to assess the viability of employing a new type of SAR. This SAR consists in the first ever Socially Assistive Drone (SAD): a quadrotor drone of reduced size conceived
for maintaining human-drone interaction during therapeutic sessions meant for children diagnosed with ASD. The main focus of this paper consists in developing an adaptive control system based on Fuzzy Logic Control due to its inherent advantages such as intuitiveness or the ease with which expert knowledge can be introduced into control systems. In total, four different behavioural modes are developed for the SAD, together with an adaptive algorithm which influences both a set of adjustable parameters associated with each mode and the likelihood of it being selected based on the user specific preferences. The four behavioural modes and their respective adaptive algorithms are tested on 10 participants in 30-minute sessions. The variations in the adaptive parameters and the likelihood of each mode being selected for each individual are registered and reveal that the SAD is able to adapt its behaviour to each participant based on their respective levels of engagement.

A hierarchical bi-level model predictive controller is proposed in this thesis to reduce the computational complexity of controlling a large scale air traffic control problem, using a model predictive control approach. The bi-level controller developed and tested in this research project is a combination of a global, long-term, slow-rate, centralized model predictive controller and a local, short-term, fast-rate decentralized model predictive controller that aims to cooperatively guide aircraft towards their destination while avoiding forbidden areas. The bi-level controller is compared to both single-level controllers it is contrived to explore the benefits achieved by the cooperation of the individual controllers, in the context of an air traffic control application. An accurate baseline model predictive controller is used to compare the computational efficiency advantage gained when using the bi-level control structure. The bi-level controller proves to attain a superior control performance over both its contributing parts. The bilevel controller provides a more accurate control performance than the single global centralized controller and performs better in trajectory optimization than the local decentralized controller. Furthermore, the controller performance of the accurate baseline controller can be approached with the bi-level controller at a reduced computation time.

Traditional electric power systems with large centralized base load power plants have a limited ability to react rapidly to the high supply variability associated with the increasing deployment of variable and intermittent renewable energy sources (RESs). Furthermore, with current power distribution networks primarily designed for unidirectional power flow, the introduction of reverse power flows by renewable feed-in strategies has been shown to negatively impact grid stability, security and system protection. For residential grid-connected microgrids (MGs) wishing to increase their renewable generation, these issues along with economic considerations often highlight that optimal operation can only be achieved through an increased self-consumption of locally generated renewable energy. Recently, researchers have highlighted that these issues can be addressed by the application of demand side management (DSM). Broadly speaking, these DSM strategies can be considered as programs which attempt to modify flexible user demands in order to achieve objectives such as reduced energy costs or increased RES utilization.

To achieve these optimal energy management goals, this thesis focuses its efforts on the application of hybrid economic model predictive control (EMPC) strategies for the DSM of small to medium sized grid-connected residential MGs, containing both local photovoltaic (PV) generation capabilities and thermal energy storage (TES) devices. In particular, the investigation exploits thermal energy storage properties of switched domestic electric water heaters (DEWHs) to optimally schedule energy demand for the minimization of MG operating costs. By considering the time varying electricity tariffs, it is shown that the implementation of EMPC is able to simultaneously target reduced electricity costs while also encouraging the self-consumption of local PV generation.

Finally, to address the unavoidable presence of uncertainty in domestic hot water (DHW) user demand, the thesis additionally explores the use of stochastic and robust variants of EMPC. By explicitly considering uncertainty, these control frameworks are able to provide more robust system operation. Specifically, the work implements min-max and stochastic scenario-based frameworks, which were shown to drastically reduce the violation of user comfort constraints when compared with their deterministic counterparts.

In this thesis, the effect of traffic light control on urban networks is investigated, with the main focus on the optimization algorithms used to solve the optimization problems resulting from a model predictive control approach. This will be done by using a macroscopic traffic flow model (the S-Model) to simulate the real-time traffic flows inside a network. A microscopic emission model (the VT-micro emission model) is further added to also include impact of the vehicle behaviour (accelerating, decelerating, driving at a constant speed, etc.) on the total amount of emission gasses that are exhausted by the vehicles inside the traffic network.

In this thesis the optimization algorithms will be investigated with respect to the reduction of the cost function as well as the computation time needed compute the (sub)optimal solution. The cost function can consist of the Total Time Spent (TTS) by the vehicles inside the network, the Total Emissions (TE) the vehicles exhaust while inside the network, or a combination of both.

We make use of Matlab for implementation of the models and the optimization algorithms and SUMO as a traffic simulator, which will be used to simulate a real-time traffic network.

The traffic flow model and the emission model are both non-smooth and non-convex, which in general would require the use of a global optimization algorithm together with multiple starting points. We also use a smoothening function on the models to work with a local optimization algorithm that works with derivatives of the cost function and both models. By using multiple starting points for this method, we hope to obtain similar results. The optimization algorithms that are implemented and investigated in this thesis are: the Genetic Algorithm (GA), the Simulated Annealing (SA) algorithm, the Pattern Search (PS) algorithm, and the Resilient backPROPagation (RPROP) algorithm.

To compare the results of the four optimization algorithms, four different scenarios are considered. The final results show that all control methods perform better than the FT controller. Overall the GA algorithm performs best without using a multi-start approach, with PS and SA having similar results. The RPROP method is either close to the other methods (0-2%) or quite far off (10-15%), depending on the scenario. When looking at computation time, PS is the fastest. It is twice as fast compared to the GA in most scenarios. SA however takes around fifteen times the amount of computation time compared to the GA. RPROP has varying results again compared to the GA algorithm. A better cost function for the TTS as well as a more optimized algorithm for the RPROP method can resolve both of these issues but future work might need to prove this.