The field of aerial manipulation has many applications, such as autonomous maintenance in difficult-to-reach or hazardous locations. However, achieving precise control of a floating-base multi-body system presents considerable challenges, particularly due to the need for robust performance under strong external disturbances in uncontrolled environments. To address these challenges, this work investigates the Incremental Nonlinear Dynamic Inversion (INDI) method as a promising control framework.

While INDI has commonly been applied to multi-rotor platforms and other single-body aerial vehicles, its use in aerial manipulation remains largely unexplored. Moreover, existing applications have focused on reference tracking in free flight, leaving the application of INDI in Aerial Physical Interaction (APhI) tasks unaddressed. To fill this research gap, we propose a novel Nonlinear Model Predictive Control (NMPC) formulation, building upon state-of-the-art works in multi-rotor control and aerial manipulation, and combine it with INDI as an inner-loop controller. Two control schemes are developed and evaluated: one in which INDI is applied solely to attitude control, and one in which INDI is used in both position and attitude control.

The aerial manipulator platform studied in this work is a Differential Shoulder Aerial Manipulator (DSAM), consisting of an underactuated quadrotor base equipped with a two-degrees-of-freedom (DoF) robotic arm.

The proposed controllers are validated through both simulation and real-world experiments, including free-flight maneuvers and APhI sliding tasks. Real-world results demonstrate that the baseline controller without an INDI inner loop is unable to successfully perform the sliding tasks, whereas both INDI-augmented schemes are able to simultaneously track end-effector position, attitude, and desired reference contact force. This is demonstrated across several trajectories on both a whiteboard using a marker and a blackboard using a crayon, as well as for simultaneous end-effector pose and force tracking in different directions in the world frame.

Finally, it has been experimentally validated that extending the controller with an INDI position control layer reduces end-effector position tracking error compared to a controller without this layer. In a real-world experiment tracking a figure-eight trajectory while simultaneously maintaining a desired contact force under external wind disturbance, the INDI-augmented controller reduced the maximum end-effector position tracking error by 45%.

THE transition to sustainable energy requires cleaner and more efficient ways to consume heat.We are used to our homes and offices being warm, yet behind this expectation lies infrastructure that must continuously balance supply and demand: heat must be produced, transported, and delivered exactly when and where it is needed. District heating networks accomplish this by circulating hot water through underground pipelines, connecting heat sources to thousands of buildings across a city. Operating such a network efficiently is far from trivial. Decisions made now about how much heat to produce, at what temperature, and through which routes have consequences that unfold over hours as hot water slowly travels through kilometers of pipe. Getting these decisions right can mean significant cost savings; getting them wrong leads to wasted energy or discomfort.
This challenge is fundamentally one of planning under complexity. The operator must anticipate future demand, account for heat losses during transport, respect physical limitations of pumps and pipes, and respond to fluctuating energy prices, all while the state of the network is only partially observable through a limited set of sensors. Traditional approaches rely on simple rules and operator experience, which suffice for conventional high-temperature networks but fall short as renewable sources have lower temperatures, requiring more advanced anticipative control strategies.
This thesis develops computational methods that enable district heating networks to be operated efficiently in real time. The underlying physics lead to mathematical optimization problems that are, in their original form, too complex to solve within practical time limits. This thesis introduces techniques that reformulate these problems into more tractable forms, decompose them across time to reduce computational burden, and learn efficient representations from operational data.We also address the question of where to place sensors to best reconstruct the network state, and establish theoretical conditions that guarantee stable closed-loop operation.
These methods are validated on realistic network models and compared against conventional rule-based strategies. The results show that predictive control can meaningfully reduce operating costs while improving constraint satisfaction, that our sensor placement method provides a practical tool for selecting informative measurement locations under tight instrumentation budgets, and that computational decomposition and data-driven surrogate models each bring solve times within real-time limits without sacrificing performance. We further find evidence of inherent stability properties of the closed-loop system, even in the absence of terminal constraints. Together, these findings demonstrate that the fundamental barriers to deploying advanced control in district heating networks can be systematically addressed.

Buildings, as major global energy consumers, can help mitigate the impact of growing renewable energy in smart grids through demand-side management (DSM). Smart energy management of buildings requires advanced control schemes that can cope with economic objectives, environmental uncertainties, occupant comfort, physical constraints, and external communication signals. Robust optimization (RO) and model predictive control (MPC) provide systematic and effective frameworks for this purpose. Beyond building energy management, RO and MPC methods are also fundamental to a broad range of engineering applications, such as chemical process planning, transportation, robotics, etc. This thesis focuses on RO and MPC design for linear systems as well as their applications in building energy management.  The research is organized into three topics:

• MPC designs for building energy management to enable energy-flexible DSM and improve environmental sustainability (Chapters 2–4).
• data-driven RO designs and algorithmic solutions for linear models to reduce conservatism and improve computational efficiency (Chapters 5 & 6).
• a distributionally robust MPC design for constrained linear systems to robustify control performance against additive model uncertainties and disturbances (Chapter 7).

Greenhouses play a critical role in the world's food production by enabling controlled crop growth in diverse and often suboptimal climates. Effective climate control is essential to maximise plant growth with minimal energy and resource usage. Natural ventilation, regulated through mechanical vents, is a key component of this control. To achieve consistent and efficient control, automation was introduced. Initially, greenhouse automation began with integrating RBC into climate computers. RBC provided a simple and structured approach to greenhouse climate control, making it easy for growers to operate the climate control system. In current practice, RBC remains widely used for the greenhouse natural ventilation control. However, over time, the accumulation of additional rules and settings has resulted in complex systems that are increasingly difficult to manage. The high number of settings and rules led to inconsistent and inefficient use by growers.

This thesis proposes a NMPC approach for greenhouse ventilation control as an alternative to the complex rules from the RBC structure. The study focuses on maintaining near-optimal greenhouse climate conditions while minimising mechanical wear and tear and reducing operational control complexity. Moreover, the thesis provides a detailed analysis of the RBC rules structure and settings of the MultiMa, a commercial climate control computer.

A simplified nonlinear greenhouse model is used to estimate the greenhouse dynamics. The model parameters are estimated with seasonal weighted offline NLLS using historical data of a commercial greenhouse. The seasonal parameter estimation creates a summer and winter model. The summer model provides an accurate estimation of the greenhouse dynamics. The modified winter model has less accurate performance, but still follows the overall trends of the measured temperature and humidity.

The models are used to simulate various NMPC approaches and compare them to the RBC system. The performance of the simulations is evaluated based on reference tracking accuracy and the total number of vent position changes, representing the ability to maintain climate conditions and minimise the wear and tear of the vents. The results of the simulation show that NMPC can achieve similar climate conditions to measured RBC climate conditions taken as reference, while significantly reducing the wear and tear. Moreover, NMPC can successfully integrate dynamic constraints, based on the current RBC constraint rules, into the optimisation.

Overall, the proposed NMPC framework presents promising results as an alternative to the complex rules and settings of the traditional control system. The \ac{NMPC} approach offers a more manageable and interpretable control system, while maintaining desired climate conditions in the greenhouse and minimising mechanical wear and tear.

Continuum soft robots present significant opportunities for advancing robotics, but they also introduce substantial technical challenges. These systems are highly nonlinear, infinite-dimensional, and severely underactuated, making control particularly difficult. While recent advancements in model-based control have addressed some of these issues for soft robotics, numerical optimal control has shown strong potential, especially given its success in other severely underactuated domains such as bipedal and quadrupedal locomotion.

However, the application of optimal control in soft robotics has largely relied on simplified models, and its use with more accurate and geometrically consistent formulations remains underexplored, particularly for explicitly tackling underactuation. This thesis investigates the use of Differential Dynamic Programming (DDP) to control continuum soft robots modeled using the Geometric Variable Strain (GVS) framework. The focus is on the Soft Inverted Pendulum (SIP) as a template system to evaluate DDP’s feasibility, robustness, and performance in underactuated settings, including low-stiffness regimes where collocated feedback strategies break down. The implementation leverages the use of analytical gradients computed via the Recursive Newton-Euler Algorithm (RNEA) to improve convergence and computational efficiency.

The results show that DDP outperforms traditional Partial Feedback Linearization (PFL) methods, both collocated and non-collocated, especially across challenging mass-stiffness combinations. This effectively extends control authority and stability into regimes previously considered difficult to handle. This thesis extends the method to more complex hybrid soft–rigid systems, examining real-time feasibility and practical implementation, thereby laying the foundation for a generalizable optimal control framework for soft robots.

The urgent need for sustainable intensification of food production, driven by a growing global population and increasing climate variability, has positioned greenhouse cultivation and advanced climate control as critical areas of innovation in agriculture. Greenhouses enable precise environmental management, significantly boosting crop yields. However, realizing these benefits requires climate and crop models that accurately represent greenhouse dynamics while remaining interpretable and suitable for real-time control.
This thesis addresses greenhouse climate prediction by developing a hybrid modeling approach combining the strengths of physics-based system identification with modern machine learning techniques. In collaboration with Hoogendoorn Growth Management, a hybrid model was developed utilizing a physics-informed Sparse Identification of Nonlinear Dynamics (SINDy) method to capture primary mechanisms influencing temperature and humidity, augmented by an Long Short-Term Memory (LSTM) neural network to account for residual, unmodeled effects. A key contribution of this research is demonstrating the feasibility of transfer learning, successfully adapting a model trained initially on simulation data to real-world greenhouse scenarios, yielding reliable predictions for both air temperature and humidity under operational conditions.
Empirical results validate that transfer learning effectively bridges simulation and practi cal greenhouse environments, underscoring the practicality of data-driven climate models for industry applications. Integrating the model into an Model Predictive Control (MPC) framework illustrates its operational viability; the controller accurately tracks temperature setpoints but experiences challenges maintaining humidity within desired limits. These findings emphasize the necessity for accurate predictive models and underscore the importance of carefully formulating MPC strategies. Further research into modeling humidity dynamics, expanding the model’s state space, and refining targeted retraining methodologies is essential to ensure robust, year-round practical deployment.
Overall, this thesis advances greenhouse climate modeling and control by showcasing how hybrid modeling combined with transfer learning can effectively close the gap between simulation based development and operational implementation, thereby laying the groundwork for adaptive and efficient greenhouse management practices

Distributed energy resources challenge the situational awareness of power flows. Many distribution grid (DG) operators have not yet implemented state estimation (SE) due to the expense or privacy constraints of measurements that lead to an unobservable system, as well as inaccurate grid parameters. A key concern with the latter is the presence of medium- and low-voltage transformers with off-load tap changers, whose tap positions critically influence voltage levels across the network. Identifying transformers where the registered position is likely incorrect — and, when on-site verification is impractical, estimating a plausible tap setting — constitutes a valuable contribution to improving network observability and operational accuracy. Although operators could manually inspect each transformer, this is impractical — there are, for instance, up to 20,000 transformers in the Southern Netherlands.

To address these challenges, this thesis proposes a novel topology-aware framework for estimating state and transformer tap positions in unobservable DGs. The framework comprises two key components. First, a generative adversarial network (GAN) is used to train a generative model conditioned on the network topology and synthetic power flow data, generating realistic measurements. Second, an integrated model, referred to as the TapSEGNN model, is proposed for estimating state and transformer tap positions.

Both of these components employ a core model architecture which combines graph and simplicial complex neural networks to capture spatial dependencies between nodes, edges, and higher-order structures. To train these components, an industrial-grade data-processing pipeline was developed using a real DG topology and simulating the exact available measurement locations. The results show that balanced adversarial training of GAN accurately imputes the missing active power injection measurements, but produces high variance in imputations for voltage magnitude and active power flow measurements. The performance of the TapSEGNN model demonstrates at least tenfold higher accuracy for SE compared to conventional methods, and it predicts transformer tap positions with 100% accuracy in a computationally efficient manner. TapSEGNN also shows promising scalability for larger networks; however, it struggles with generalisability across similar real networks. Finally, the suboptimal performance of both components in certain aspects warrants further investigation, which is recommended as future work.

In recent years, the yearly share of electricity generated in the Netherlands from renewable sources such as solar and wind has increased significantly, reaching 42% in 2023, and is projected to rise to 70% by 2030. However, production from these sources is highly weather-dependent, unpredictable, and often misaligned with typical periods of peak electricity demand. This mismatch leads to price volatility in electricity markets, emphasizing the importance of a reliable trading strategy for assets with flexible production or demand. Solar panels are partially flexible assets, as their production can be curtailed. However, their output is weather-dependent and cannot be predicted with complete accuracy. This creates a challenge in determining the trade volume in the day-ahead market, resulting in the highest profit. Trading conservatively in the day-ahead market reduces revenue but minimizes imbalance volumes through the possibility of curtailment. On the other hand, trading too much can result in unavoidable imbalances when actual production falls short. Existing literature focuses on minimizing imbalance volumes as extreme prices, high volatility, and lower average prices than day-ahead prices characterize the imbalance market. In these works, it is typically assumed that the real-time imbalance price is unavailable. However, real-time imbalance price predictions are available in this research, enabling optimal real-time decision-making. This provides the opportunity to profit from high imbalance prices while avoiding negative prices. In this study, a coordinated bidding strategy optimizes day-ahead bids, balancing revenue maximization and risk minimization as both the production and imbalance price are uncertain. The primary objective is to explore methods for incorporating uncertain imbalance prices into day-ahead optimization. Various methods from the literature are compared, and a novel decision-focused approach is introduced. Combined with solar generation forecasts, state-of-the-art day-ahead price predictions, and optimization models, monthly revenues are simulated using \ac{EMS} software. Results show that modeling uncertain imbalance prices using historical scenarios achieves the highest and most consistent revenues, especially when combined with \ac{CVaR} optimization. The novel decision-focused approach also performs among the best models, delivering consistently high revenues. Adding battery storage to the solar panels yields similar results, and further revenue increases are possible with improved solar forecasting. This highlights an important direction for future research.

This thesis introduces two improvements in trajectory planning on uneven terrain with dynamic obstacles: the Dynamic Traversability-based Hybrid A* (DT Hybrid A*) algorithm and the Safety Direction Velocity Obstacle (DSVO) replanning method. The DT Hybrid A* differs from traditional path-planning approaches by adopting a less conservative strategy, incorporating vehicle momentum into terrain traversability in order to more accurately reflect the vehicle’s capability to navigate uneven terrain. Based on dynamic traversability, the DT Hybrid A* is developed by leveraging a kinematic bicycle model for steering, allowing variable travel lengths between adjacent nodes, and incorporating both distance and time heuristics. These enhancements significantly reduce CPU time costs during the planning process, resulting in shorter path lengths and faster planning times. This thesis incorporates a collision detection mechanism within the global path planning framework, enabling the system to replan if a potential collision is predicted in the future. The replanning process initializes the global path to avoid all static obstacles. With the movement of dynamic obstacles, if a potential conflict is recognized, the vehicle discards the current global path and replans from its current location. To ensure that the replanned path is free of dynamic obstacles, the collision detection mechanism incorporates Safety and Direction Elements based on a Velocity Obstacle (VO) cost term into the global path algorithm (DSVO). The DSVO algorithm enhances collision avoidance by integrating Safety and Direction Elements based on a VO cost term, surpassing traditional methods that use Euclidean distance penalties. DSVO avoids collisions by considering obstacle velocity and direction over time. It enhances the rate of successful replans for dynamic obstacle avoidance, producing the shortest collision-free paths with minimal computational overhead and completion time. The effectiveness of these approaches highlights the balance between computational efficiency and dynamic obstacle management, paving the way for more reliable path replanning solutions. DSVO surpasses methods that use Euclidean distance as a penalty for dynamic obstacles because it considers the obstacle’s velocity and direction over time. It also outperforms the Safety Element based on a VO cost term, which only ensures collision avoidance within a short time frame and provides the optimal velocity for that period but does not account for longer-term safety, potentially leading to collisions over extended durations. Additionally, since these are cost terms within global path planning algorithms, there can be conflicts between obstacle avoidance and achieving the goal. The DSVO algorithm, which incorporates Safety and Direction Elements based on a VO cost term, mitigates conflicts between obstacle avoidance and goal achievement. Additionally, DSVO effectively adjusts the path to navigate around dynamic obstacles while maintaining a direct route toward the goal, balancing safety and efficiency.

This thesis presents the design and implementation of a process monitoring system for the HIsarna pilot plant aimed at the early detection of anomalies, specifically foamers. Though rare, foamers represent severe process faults that can significantly impact the plant’s safety and reliability. Early detection of these events is crucial for maintaining operational stability.

Given the ambiguous definitions of nominal and non-nominal operations in HIsarna, this thesis constructs a precise definition based on foamer characteristics. Foamers are non-nominal operations, while the absence of these foamer characteristics indicates nominal operations. The monitoring system utilizes data from a gas analysis sensor in the plant’s topside outlet, the dogleg. The dogleg gas analyser is chosen for its reliability and the absence of monitoring methods based on this sensor. To ensure the accuracy of the sensor data, an optimizationbased method is employed to eliminate air ingress disturbances.

An anomaly detection method based on distance measures is developed, incorporating the HIsarna Operation Model (HIOM) to reduce sensitivity to operating point changes. Historical process data is analyzed to establish a distance measure that reflects the correspondence of current measurements to nominal operations. The performance of the anomaly detector is evaluated based on the number of foamers detected, hours per false alarm, and detection time.

The primary contribution of this thesis is the development of an anomaly detection algorithm using Mahalanobis distance for the HIsarna pilot plant. This metric indicates how closely current measurements match historical nominal operation data. Additional contributions include a rigorous definition of nominal and foaming operations, an optimization-based method for air ingress removal, and an observer for dogleg gas analysis data, which can estimate carbon conversion and oxygen flow prediction bias of HIOM.

Model matching control allows for a virtual change in the vehicle's physical parameters, removing the need to alter the system physically. These vehicles are used as a versatile test bed for safe and repeatable experiments, and allow for safer vehicle operation by virtually keeping the driving behaviour constant in the face of physical changes such as wet roads. Although already applied on planes and cars, it has not been applied to bicycles. The proposed control method accomplishes model matching control by equating the state derivative of the physical system to that of the the virtual system. Theoretical analysis shows model matching control is applicable to bicycle dynamics, even if only the steer torque is available for control. Via experiments, I identify the bicycle's weave mode eigenvalue at different speeds, and the Bode gain between steer torque to lean rate and to fork angle for frequencies between one and three Hz. The experimentally identified values follow the theoretically predicted values closely but do not overlap perfectly. The effect of artifacts during the experiment and unmodelled dynamics are analysed with the use of simulations and are able to explain these perceived differences. From the results, this research concludes that model matching control is applicable to bicycles and is able to virtually change the bicycle's physical parameters. Future work will look into making the controller more robust to modelling errors.

Falling remains a large source of (traumatic) injuries and healthcare costs. Over the past years, different actuators have been developed in the field of wearable robotics to help prevent injuries from falling. To increase the wearability of these systems, the weight of power storage can be decreased with intermittent instead of continuous control. A fall detector is then needed for these systems to trigger the activation of the actuator to prevent the fall. A proof of concept for a model-based fall detector that is aimed at using only wearable sensor measurements is presented. The algorithm is based on a single inertial measurement unit placed on the lower back. The upper-body orientation and centre of mass velocity are estimated with two separate Kalman filters. The velocity is estimated with a gait model consisting of a spring-damper-legged point mass in three-dimensional space. The balance of the subject is evaluated with the velocity estimates and the extrapolated centre of mass method. The presented model is verified on a non-falling treadmill walking dataset of real humans and shows accurate estimates of the centre of mass velocity. Furthermore, a planar falling simulation is performed to show successful pre-impact fall detection. This resulted in no false alarms outside the initial estimation settling time, and a successful detection with a lead time of 680 ms. This lead time is long enough to provide a trigger for fall mitigation devices.

In light of the pressing challenges posed by global climate change and the imperative to reduce CO2 emissions, innovative approaches in energy management are critically important. This thesis presents an exploration of heat pumps integrated with Thermal Storage System (TES) systems, an area of research and application pivotal for enhancing energy efficiency and environmental sustainability. The combination of heat pumps and TES systems emerges as a key factor in reducing greenhouse gas emissions and optimizing the utilization of renewable energy. Such integration plays a crucial role in minimizing operational costs, reducing environmental negative impact, and augmenting system efficiency by enabling the storage and later use of energy from renewable sources. Moreover, this integration facilitates the effective management of demand-side energy, bolstering the capacity to incorporate fluctuating renewable generation into the energy grid. This is achieved by dynamically load shifting to balance energy supply and demand.

A central aspect of this thesis is the utilization of Model Predictive Control (MPC) for advanced energy management. The research delves into the use of MPC to optimize the operational economy of the system, aiming to maximize cost-efficiency. Additionally, an innovative MPC-based Demand-Side Management (DSM) strategy is introduced. This strategy involves two key steps: initially establishing a model to assess the system's energy flexibility, followed by harnessing this flexibility to respond to demand fluctuations. Such an approach facilitates dynamic adaptation to varying energy demands, ensuring optimal resource utilization. The predictive capability of MPC, which accounts for future disturbances including demand forecasts, electricity pricing, and weather conditions, is exploited to improve the system’s responsiveness and operational efficiency.

Experimentation was conducted both in simulations and through the implementation in real systems. These practical applications demonstrated significant savings in energy costs and energy consumption, achieving economical operation. Furthermore, the execution of the proposed two-step demand-side management strategy successfully managed energy demands. This not only underscores the practical effectiveness of the proposed system but also highlights its potential in real-world scenarios.

In summary, this research underscores how the integration of heat pumps, TES systems, and advanced control strategies like MPC can significantly improve energy efficiency, reduce operational costs, and enhance energy flexibility. It highlights the vital role of incorporating sophisticated control mechanisms into sustainable energy systems, aligning with the strategic goals of modern energy policies and advancing the field of sustainable energy management.

Greenhouses offer the promise to mitigate the challenges faced by traditional open field agri- culture. Operating these systems on a commercial scale demands effective control and forecast models. This thesis contributes to the increasing field of research that applies methods from systems and control to greenhouse systems. The primary objective was to evaluate various cli- mate prediction models for their applicability in forecast and predictive control within green- houses. Comparative analyses were conducted on linear and non-linear data-driven models across one week and one hour time horizons. The linear models include an autoregressive with exogenous input (ARX) model, subspace identification using the Multivariable Output-Error State-space algorithm (MOESP), and dynamic mode decomposition with control (DMDc). Of the linear methods, MOESP performed the best on both time horizons. The nonlinear methods included several long-short term memory (LSTM) models with different architec- tures. A stacked LSTM model was found to outperform the other LSTM models as well as all linear models over the one-week time horizon. The Koopman with inputs and control (KIC) framework, a novel nonlinear approach, was also examined. The choice of lifting func- tions proved to be a challenge, yielding only partial improvements over linear methods on the training set. For the short-term one-hour time horizon, the data-enabled predictive control (DeePC) algorithm was also implemented and yielded the most accurate results. Notably, the stacked LSTM predictor did not show any advantages on the one-hour timescale. Finally, a novel approach combining Koopman theory with DeePC was introduced and compared to the regular DeePC algorithm. No improvement was observed with this proposed method for one hour-ahead predictions. The comparison concluded with the insight that for the one hour time-horizon, the nonlinear methods that were investigated did not show advantages over linear methods in the greenhouse system. On the one week time horizon, nonlinear methods outperform linear ones. It affirmed the challenges around finding observables that fully span a Koopman invariant subspace. Further investigation on the proposed extension to DeePC should be the focus of future work.

A flexible manufacturing system (FMS) has advantages over traditional manufacturing systems due to its ability to deal with unpredicted circumstances such as changes in demand or component breakdowns by re-routing. However, this flexibility increases the complexity of controlling such a system. Traditionally, the system model is simplified to reduce the solution space by removing intra-machine transportation complexities. This thesis explores how these complexities can be kept and accounted for during scheduling. A scheme is used where short term schedules are continuously calculated to determine the optimal schedule over the next timeframe. The flexible job shop scheduling problem with transport (FJSPT) is used to represent the complexities of the FMS. To calculate part-schedules repeatedly a fast constructive search method is needed, the AlphaZero framework is identified as a fitting candidate. The FJSPT is translated into the reinforcement learning framework using a reduced action space, a graph neural network based state representation and normalized reward function. A naive normalization approach for the reward function is found to introduce problems in the value function sensitivity, while other adaptive method show fundamental flaws. A novel normalization method is introduced using min-max adaptive normalization and suboptimal node inclusion to improve value function training data. Implementing and training the algorithm shows the method performs poorly in comparison to metaheuristic based algorithms for the FJSPT problem. The value function is not able to converge to training data, while this is critical for the self-improvement training of the algorithm. Future work should focus on developing a normalized value function that is sensitive to solution quality and is able to converge. Despite the challenges, the work provides insights into the complexities of implementing AlphaZero for combinatorial optimization.

Robot manipulators are significantly more accurate than their human counterparts and enhance the repeatability of various tasks. However, manufactures still provide reduced information regarding the robot controller functions, typically affecting robots' predictability. Additionally, industrial examples show that system transparency could be improved, helping users to understand the process and apply corrective actions. One step in the direction of improved predictability is the identification of the Limiter block associated with the KUKA robot controllers that limit the acceleration and jerk obtained during motions. This work aims to provide dynamically feasible trajectories with reduced performance deterioration. Specifically, the Limiter's behaviour was analyzed and later predicted in a iterative manner, aiming for the maximum velocity, acceleration and jerk that can be achieved, starting from the current robot state and given specific robot information. Using the predicted bounds, the trajectory generator will determine the maximum achievable point that fulfills all the constraints, running an optimization problem in an iterative fashion. Nonetheless, the practical experiments represent a significant component allowing data collection and results validation.

Humanitarian, environmental, and political concerns have contributed to the evolution of agricultural technology, also known as AgTech. Researchers from various scientific backgrounds are diving into AgTech to ensure the world's food security, create a sustainable future for agriculture, and pave the way for autonomous cultivation methods. This thesis project attempts to contribute to the aforementioned subjects from the control engineering perspective. The thesis objective is the design and testing of a novel predictive climate controller, for tomato greenhouses agriculture, able to conclude on the optimal yield-energy consumption ratio with limited intervention by the human factor. The main novelty introduced by this algorithm is the use of crop variables in the decision-making process according to the Speaking Plant Approach (SPA). However, no straightforward recipe indicates which crop signals could be used. In the context of this study, it is explored how crop variables, measurable by thermal imaging, can be used for the formulation of a SPA-based objective function. Specifically, the research focuses on stomatal conductance, the canopy, and the mean canopy temperature, for the SPA-based objective function formulation. The cost function generation entails the definition of the necessary state constraints. Except for the objective function definitions and the determination of constraints, a predictive controller requires a system representation that acts as a predictor. Nevertheless, the complex and non-linear nature of the climate-crop system complicates the system identification process. Concurrently, data science is blooming and new data-driven system representation techniques are breaching. Data-Enabled Predictive Control is a novel control policy based on systems behavioral theory which uses a non-parametric system representation enabling the omission of the system identification process. This approach has not been tested for the description of highly complex climate-crop systems. Therefore, another target of this project is to examine the capabilities of this data-driven predictor for the representation of the climate-crop model and evaluate if and how can be used as a predictor in a climate control regime.

Creating autonomous Micro Aerial Vehicles for executing complex missions poses various challenges, including safe navigation in the presence of external wind disturbances. Most current navigation methods handle external wind disturbances through real-time estimation and rejection algorithms in the control stage, but lack safety guarantees in strong winds. Recent robust methods provide safety guarantees but can be overly conservative. With the availability of more powerful computing devices, data-driven control algorithms are becoming increasingly feasible. Combining Gaussian Process models with Model Predictive Control has shown to enhance safety and performance in various control applications. Moreover, Model Predictive Control has been extended to solve more complex optimization algorithms that combine trajectory generation and tracking, preventing reference trajectories that are risky and challenging to control in the presence of disturbances.

This research aims to improve quadrotor navigation in windy environments by using Gaussian Processes to model wind disturbances. The Gaussian Process model is integrated with a Model Predictive Contouring Control formulation that combines trajectory generation and control into one optimization problem. A nominal model of the quadrotor is derived and the Gaussian Process disturbance model is trained with the quadrotor position as input and wind disturbance as output. The wind disturbance map, along with the nominal model, is implemented in the Model Predictive Contouring Controller to consider both the mean and uncertainty of the resulting probabilistic model.

This study validates the use of Gaussian Processes to model complex wind disturbances in quadrotor navigation. The wind disturbance map is trained from available state information, with data collected using an optimal exploration design to minimize uncertainty and reduce exploration times. The hyperparameters involved in training the Gaussian Process model are discussed and the implementation of sparse Gaussian Processes is outlined to make it real-time feasible for the Model Predictive Contouring Control formulation. Including the prediction model by incorporating chance constraints results in improved tracking and increased robustness in cluttered environments compared to the nominal Model Predictive Contouring Control formulation. The proposed algorithm is shown to outperform state-of-the-art methods for safe quadrotor navigation in windy and cluttered environments, being able to handle more complex wind fields than existing methods while also being less conservative.