To achieve technical compliance with the Specific Operations Risk Assessment (SORA) framework is a significant challenge for Beyond Visual Line of Sight (BVLOS) drone operations. A reliable Command, Control, and Communication (C3) Data Link that meets SORA regulations is crucial for integrating drones safely in sectors that face high risk and labour shortages, such as inspection of long power transmission lines and wind turbines. In this work, we focus on C2 design aspects of an in-house data link tailored to BVLOS drone operations. Central to our design approach is the operational and mission capability diagram, which bridges the practical operational needs with key SORA objectives.

We implement a systematic process to translate relevant SORA objectives into quantifiable technical requirements, referencing 3GPP standards and JARUS guidelines. Subsequently, we design a sequence diagram mapping C2 data link functions to operational scenarios, which transforms into a logical architecture supporting modular Models-Based Systems Engineering (MBSE).

We perform test and results show data received by the Ground Control Station (GCS) server increased by threefold while latency reduced by 92\% by making improvements to the Scheduler and Link Manager modules respectively. In addition, we design a Link Manager simulation to offer clear metric visualisation to enable more confidence in operator assessment before physical flight trials. We show using simulations and experiments the benefits of our design. Our proposed framework enables wider drone deployment for BVLOS missions in the Netherlands and across Europe.

This thesis asks whether spiking neural networks (SNNs) and neuromorphic computing constitute a promising alternative to present-day artificial neural networks (ANNs) for autonomous space missions. Focusing on a resource- and power-constrained 1U CubeSat transiting the Van Allen radiation belts, TIENOS is a toolchain that injects radiation-inspired perturbations into trained models and records the layer-specific reactions.

The framework systematically emulates dominant soft-error mechanisms by applying (i) bit-flip faults representative of single-event upsets, (ii) additive Gaussian noise as a proxy for thermal/analog variability, and (iii) dropout-style masking to approximate transient loss or zeroing of activations. Using MNIST (frame-based) and N-MNIST (event-based) benchmarks, we compare LeNet-5–style convolutional neural networks and size-matched multilayer perceptrons with their spiking counterparts to establish an ideal software training baseline. The tool produces per-layer vulnerability profiles and robustness heatmaps across a broad range of perturbation rates, quantifies activity sparsity in SNNs, and can be used to evaluate noise-aware retraining to improve robustness without any overhead, with a path towards on-chip protections such as selective redundancy (such as triple-modular redundancy for neuron parameters), ECC and scrubbing.

Results show that fragility concentrates in a limited subset of layers depending on the fault mechanism, enabling targeted hardening with modest cost. It also indicates that noise-aware retraining improves tolerance without prohibitive accuracy loss and that SNN sparsity yields favourable energy–robustness trade-offs for bursty, event-driven sensing typical of small spacecraft. In this study, the noise-aware learned weights used for inference by the two-tinyODIN setup deliver a 15% higher accuracy for up to 2% of bit-flips in the system. Hybrid ANN–SNN pipelines could further enlarge this envelope by deploying spiking computation where sparsity is highest while retaining dense processing elsewhere, acknowledging that the scalability and baseline power of current neuromorphic platforms remain practical constraints. Overall, the methodology translates environmental assumptions for a 1U CubeSat in the Van Allen belts into actionable, layer-level design rules, providing a principled basis for space-grade, energy-efficient digital SNN accelerators and an open, extensible tool to localise and mitigate radiation-induced vulnerabilities.

Accurate cooperative-localization of stationary agents and tracking of mobile targets are critical for multi-agent autonomy, particularly in global navigation satellite system (GNSS)-denied environments such as maritime search-and-rescue (SAR) missions. In such settings, agents often lack reliable global positioning and complete target observability, challenging distributed perception and coordination. State-of-the-art approaches such as joint Kalman filtering was widely applied.

To address this, we propose two approaches: a sequential optimization strategy and a unified integrated optimization framework. The sequential method decouples localization and tracking—first estimating agent positions via inter-agent ranging and then performing distributed Gaussian Process (GP) tracking using alternating direction method of multipliers (ADMM) -based fusion. Although efficient and modular, this approach may suffer from error propagation. To mitigate this, we introduce integrated optimization framework that couples both tasks via a weighted multi-objective cost. A convex relaxation of the localization subproblem yields closed-form updates, and the full problem is solved in a distributed manner using ADMM.

Simulations based on GNSS-denied maritime scenarios show that both methods enhance tracking and localization performance, with the integrated framework offering superior speed. These results underscore the value of cooperative self-localization and distributed tracking in complex multi-agent environments.

The usage of robots replacing human tasks has become more prevalent. Controlling multiple of these robots can be useful in applications such as disaster response, surveillance and exploration. This form of control is often achieved by using geometric patterns, such as triangles, squares, etc. Drones can employ this concept of formation control to fly and maneuver through environments and obstacles.
In this work a distributed affine formation control algorithm is implemented onto a Crazyflie drones from Bitcraze. Ultra-wideband is used for positioning and communication between the drones. The implementation of the affine formation control algorithm is optimised such that it is only executed when new information is available, to prevent onboard microcontroller from bottlenecking. This resulted in the drones flying in formation successfully with an accuracy of approximately 6.80 cm from its expected position.
Additionally, this algorithm is extended to manage cases where unexpected missing drones could comprise the stability of the formation. The implementation uses the CMSIS library that is optimised for matrix operations. This resulted in the drones flying in formation successfully even in the case of an observation loss with an accuracy of approximately 17.69 cm.
This work not only provides empirical data of experiments with an affine formation control algorithm, but also provides a baseline implementation for future research in the field of affine formation control, which can potentially lead to noise analysis or the application affine formation controls under different circumstances.

Autonomous navigation is a critical aspect of robotic systems, particularly in hostile and uncertain environments, and robot localization is central to navigation. Robot localization establishes its position within its surroundings. This thesis addresses the challenge of robot localization, focusing on a lunar-like environment with constraints such as limited computational resources and the use of non-visual based sensors. In this thesis, three sensors — wheel encoders (WE), Sun sensor (SS), and inertial measurement unit (IMU) — are employed for localization. Each sensor contributes distinct information regarding position and orientation. However, individual sensor measurements suffer from inherent inaccuracies and errors, especially the IMU’s reliance on integration over time, leading to significant drift. To mitigate these challenges, three fusion methods are explored: sensor selection based on predefined thresholds, Kalman filtering, and weighted fusion. Results indi- cate substantial improvements in localization accuracy compared to individual sensor measurements. The weighted fusion method, in particular, demonstrates superior per- formance by assigning appropriate importance, according to their accuracy, to each sensor’s information, resulting in significantly reduced positioning errors. The maxi- mum localization error using this method is 92m, which is smaller than reported in the literature. Further, the maximum localization percentage error over 65m is around 8%, which is comparable to the literature with visual sensors. The weighted fusion method introduces only a marginal increase in the computational complexity. Thus, this method stands out for its simplicity and delivers results superior to those documented in existing literature for non-visual sensors. Despite promising results, the research is met with certain hurdles, notably the avail- ability and consistency of datasets. The reliance on existing datasets, such as the Devon Island Rover dataset, highlights the need for standardized and comprehensive datasets for thorough testing and validation. Calibration inconsistencies and verification issues further underscore the complexity of real-world implementation. Nevertheless, the findings of this thesis offer insights into the integration of multiple sensors for enhanced localization in lunar-like environments. By leveraging complementary sensor data and employing efficient fusion techniques, the proposed approach enables more accurate and reliable navigation of lunar micro rovers, thus advancing the capabilities of autonomous robotic systems for future lunar exploration missions.

Climate change poses a serious threat to ecosystems and increases the need for accurate and rigorous monitoring of ecosystems. Current monitoring solutions are often bulky, expensive, and lack critical functionalities such as on-board inference capabilities, robust wireless connections, and a diverse sensor suite. Ecological monitoring projects often suffer from inefficiencies caused by the large time delays between collecting data and analyzing said data, as well as having to spend large amounts of time in the field setting up the sensors manually. This thesis addresses many of these issues by designing a sensor with an extensive sensor suite, robust wireless capabilities and an on-board audio classifier able to perform real-time inference. Furthermore, attention is paid to making the system extendable in the future and allow for potentially integrating the sensors with a drone delivery- and retrieval system. The system tests performed indicate that the system has great potential given more time to tweak some of its identified shortcomings.

This thesis presents an approach to monocular depth estimation for Unmanned Aerial Vehicles (UAVs). Monocular depth estimation is a critical perception task for UAVs, enabling them to infer depth information from visual data without relying on heavy or power-consuming sensors such as LiDAR or stereo cameras. Given the operational constraints of UAVs, such as limited payload and energy resources, robust and efficient depth estimation methods are required to facilitate safe navigation and environmental interaction. The proposed methodology in this thesis integrates visual data from a monocular camera with inertial measurements from an Inertial Measurement Unit (IMU) sensor. This combination aims to address challenges such as scale ambiguity in the depth estimates and inaccuracies in dynamic environments that are common in aerial operations. The integration of IMU data with a differentiable camera-centric Extended Kalman Filter (EKF) allows for better ego-motion estimation, effectively calibrating the visual information with drone dynamics. The method further incorporates depth map frame prediction, leveraging initial depth estimates along with temporal dynamics to predict future depth maps. This predictive capability improves efficiency by reducing the need for full depth estimation in every frame, allowing robotic agents to anticipate environmental changes. The evaluation on simulated and real-world datasets shows that while the algorithm performs well over short forecast horizons, accumulating errors from IMU data and the assumption of a static environment limit its long-term accuracy. The future depth map prediction algorithm reduced the need for DynaDepth from 10 runs per second to 2, and on the Mid-Air dataset, from 25 to 5. Additionally, this study provides a foundation for future work, including the integration of an object-oriented frame prediction algorithm.

Traditional target tracking using monostatic radar systems typically relies on centralized or decentralized architectures, where all data is transmitted to a fusion center for processing the position and velocity of mobile targets. This approach not only introduces a single point of failure and can lead to increased data transmission times, particularly when the fusion center is distant from individual radar nodes, but it also faces scalability issues and potential bottlenecks when data accumulates at the fusion center. To address these challenges, we introduce a Distributed Alternating Direction Method of Multipliers (DADMM) for target localization within a radar network, wherein each radar node shares its observed data only with immediate neighboring nodes, achieving consensus on the estimated target locations and velocities. Our simulations, which incorporate critical parameters such as the number of radar nodes, radar geometry, and Signal-to-Noise Ratio (SNR), assess their impact on estimation accuracy and convergence speed. The results demonstrate that the proposed DADMM not only effectively eliminates the single point of failure, but also enhances system efficiency and robustness. We also incorporate two distinct stopping criteria for position and velocity estimations, enabling us to promptly fix the first accurately estimated parameter and reallocate computational resources to more effectively refine the remaining parameters, which streamlines computational efforts by focusing on unresolved parameters. We highlight the additional benefits of our proposed framework, and present directions for future work.

In autonomous driving, environmental perception, crucial for navigation and decision-making, depends on integrating data from multiple sensors like cameras and LiDAR. Camera-LiDAR fusion combines detailed imagery with precise depth, improving environmental awareness. Effective data fusion requires accurate extrinsic calibration to align camera and LiDAR data under one coordinate system. We aim to calibrate the camera and LiDAR extrinsic automatically and without specific targets. Targetless, non-automated calibration methods are time-consuming and labor-intensive. Existing advanced methods have proven that automatic calibration methods based on edge features are effective, and most focus on the extraction and matching of single features. The proposed method matches 2D edges from LiDAR's multi-attribute density map with image-derived intensity gradient and semantic edges, facilitating 2D-2D edge registration. We innovate by incorporating semantic feature and addressing random initial setting through the PnP problem of centroid pairs, enhancing the convergence of the objective function. We introduce a weighted multi-frame averaging technique, considering frame correlation and semantic importance, for smoother calibration. Tested on the KITTI dataset, it surpasses four current methods in single-frame tests and shows more robustness in multi-frame tests than MulFEAT.
Our algorithm leverages semantic information for extrinsic calibration, striking a balance between network complexity and robustness. Future enhancements may include using machine learning to convert sparse matrices to dense formats for improved optimization efficiency.

The Constant Tone Extension (CTE) function introduced since Bluetooth Low Energy 5.1 greatly improves indoor Bluetooth Low Energy localization. These small, low-energy beacons transmit signals that Bluetooth-enabled devices can use to calculate proximity and positioning. This technology enables continuous navigation, location-based services, indoor wayfinding, and so on. The new feature, used together with an appropriate antenna array, can calculate the Angle of Arrival (AoA) and Angle of Departure (AoD). This study is solely concerned with AoA mode analysis.

Estimating AoA is challenging, especially indoors. Multipath signals, which greatly corrupt results in such a complex environment, cannot be ignored. The low-energy equipment used in BLE frequently results in a substantial frequency offset, which cannot be overlooked. Several elements are thoroughly examined in order to comprehend the positioning algorithm.

The processing of I/Q data is the first step in the prokect. Then we look at how frequency offset is created and how it affects estimation. We use maximum likelihood to calculate frequency offset after modeling the data structure. Following that, numerous AoA estimation and multipath-solving methods are discussed. We discuss the operation of a 4x4 URA and the advantages and disadvantages of having multiple antennas. A virtual antenna (VA) solution addresses hardware limitations, but it fails when it comes to multipath. Finally, we model AOA estimation and use the estimated angles to localize. The Matlab algorithm, as well as the LS and TLS methods, are introduced.

In this paper, we propose an end-to-end indoor BLE positioning solution. The algorithm is tested using a Matlab simulation. The multipath scenario is created by simulating an empty room with raytracing and focusing on first-order reflection routes. The simulation demonstrates that Toeplitz Reconstruction (TR) works best with the appropriate settings. Over 90% of the results in a 4-by-4 uniform rectangular array (URA) have position errors of less than 0.14 meters. In about 60% of cases, increasing the size of the 8-by-8 Uniform Rectangular Array (URA) results in a position accuracy of less than 0.1 meters.

Following the simulation, a real-world experiment is conducted to assess the solutions' practicability and efficacy. To reduce uncontrolled multipath interference, the experiment was carried out outside. The TR method has a distance inaccuracy of 0.4 m, whereas conventional methods have a distance inaccuracy of 0.1-0.2 m. The most difficult challenge is estimating elevation angles, which necessitates additional research. The results of azimuth angle estimation studies match those of 1-D AoA estimation studies. Even with a high prediction error for elevation angle, the distance error is lower than in previous positioning research.

Finally, we make recommendations for future research. This could include optimizing algorithm parameters, taking antenna-related factors into account, changing CTE configurations, and so on.

In an era marked by the demand for unprecedented levels of precision in engineering applications, the profound impact of friction forces on motion control systems cannot be underestimated. This thesis extensively investigates the frictional behavior of the Proton Motion Stage, an advanced high-precision motion control system developed by Prodrive Technologies. This research conducts both experimental investigations and computational simulations, offering valuable insights into its friction behavior across diverse conditions and scenarios.
The research begins with an analysis of existing models used to describe friction behavior in precision engineering systems. A critical evaluation of empirical models highlighting strengths and limitations is presented, and the LuGre friction model is selected. Subsequently, a simulation work is conducted to identify the viscous coefficients, the stiffness coefficient, the Coulomb friction, the Stribeck friction, and the Stribeck velocity in the LuGre model. The simulation setup is described, including the incorporation of the LuGre friction model and the identification of system parameters. The accuracy of the identification value to the true value is above 99\%. A comparison of the sensitivity of the objective function to the change of parameters is also conducted to enable a comprehensive exploration of friction dynamics. Finally, the research delves into static and dynamic parameter experiments, where cable slab forces' position-dependent impacts and velocity-friction maps that capture the intricate Stribeck effect are presented, and closed-loop and open-loop setups to dissect friction behavior during rapid motion changes are employed. Residual analysis of histogram and 90\% confidence autocorrelation and cross-correlation is also presented to study the quality of identification and shows that the LuGre model does not fully capture the friction phenomena on the Proton Motion Stage. Future research should involve the modification of the LuGre model and data-driven approaches such as machine learning. Overall, this thesis fills the gap in state-of-the-art works by combining theory and practice to enhance the understanding of friction in precision engineering systems.

This thesis project has been done with Crownstone, a subsidiary company of Almende. One of their project goals was to develop indoor localization algorithms to determine room-level location of an asset within the Crownstone network for smart-building, home automation and healthcare applications. An asset is a wireless device transmitting Bluetooth messages that are heard by Crownstones(sensors) and measure the strength of the received signal(RSSI). The terms position and location are used to denote two different concepts in this thesis. Position refers to estimating the specific coordinates, whereas a location refers to a much wider space like a room. In this thesis, we are interested in getting a location estimate.
There are two most widely used and researched localization techniques to determine the location of the asset. First, a model based(MB) method (e.g. Trilateration algorithm) which uses a mathematical model based on distance and second is a data-driven(DD) method (e.g. Fingerprinting algorithm) that relies on existing data, like RSSI to directly get the location. Algorithms are tested on real data collected by the Crownstones at the Almende office(test environment) divided into a finite number of locations or rooms. Metrics are defined based on the requirements of Almende to compare the MB algorithms with the DD algorithms. In this thesis, firstly, a centralized multilateration(MB-C) algorithm is implemented taking into account distances from N Crownstones at the office. Since one of the requirements was to perform in-network localization, a simple averaging consensus based distributed(MB-D) algorithm was selected and compared against the MB-C algorithm. Results show that the MB-D algorithm is faster, scalable and robust against single-point of failure than the MB-C but is less accurate and does not converge to the centralized solution for a noise variance greater than 10dB.
The MB algorithms have limitations in terms of selecting a model, learning the
model parameters and an additional step of mapping the position output of the implemented MB algorithms to a location is also required. To deal with these challenges, a Machine-learning(ML) based data-driven algorithm is proposed. In this, training datasets were iteratively improved with different features. Then, an Ensemble based centralized ML algorithm (DD-C) is implemented, giving a classification accuracy of 65%. Algorithm is further improved by distributed data handling leading to a classification accuracy of 77%. There has been very little to no study on finding the room-level location of an asset in an indoor setting using a distributed ML based data-driven algorithm. A consensus based distributed ML algorithm (DD-D) is proposed that performs local predictions within the Crownstone network using the same globally trained model giving a classification accuracy of 73%.
The results show that the proposed DD algorithms perform better than the MB
algorithms in terms of accuracy and are comparable in terms of prediction time. Results also indicate the proposed DD algorithms are more scalable, robust against noise but are computationally expensive.

Gaussian process regression (GPR), a potent non-parametric data modeling tool, has gained attention but is hindered by its high com- putational load. State-of-the-art low-rank approximations like struc- tured kernel interpolation (SKI)-based methods offer efficiency, yet lack a strategy for determining the number of grid points, a pivotal factor impacting accuracy and efficiency. In this thesis, we tackle this challenge.
We explore existing low-rank approximations that facilitates the computation, dissecting their strengths and limitations, particularly SKI-based methods. Subsequently, we introduce a novel approxima- tion framework, MKISSGP, which dynamically adjusts grid points us- ing a new hyperparameter of the model: density, according to changes in the kernel hyperparameters in each training iteration.
MKISSGP exhibited consistent error levels in the reconstruction of the kernel matrix, irrespective of changes in hyperparameters. This robust performance forms the bedrock for achieving accurate approx- imations of kernel matrix-related terms. When employing our rec- ommended density value (i.e., 2.7), MKISSGP achieved a comparable level of precision to that of precise GPR, while requiring only 52% of the time compared to the current state-of-the-art method.

The evolution of aerial vehicle technology necessitates robust trajectory prediction models. These models are crucial for maintaining safe airspace and enabling autonomous operations. Automatic dependent surveillance–broadcast (ADS-B) is a surveillance system that enables aircraft to receive data from navigation satellites and periodically broadcasts it, enabling it to be tracked. Moreover, using ADS-B data for more general aerial vehicles has become a popular trend because it can provide real-time high-resolution aircraft state information and share this information with other vehicles in real-time for the aviation safety ecosystem.

In this project, we delve into ADS-B-based trajectory prediction for both aircraft and drone motion trajectories with the overarching goal of improving prediction accuracy. We initially implement several model-based Kalman filters—including interactive multiple models (IMM)—to assess the accuracy of aircraft trajectory predictions across different model structures. The results reveal that the IMM filter outperforms the single model predictions in terms of root mean square error (RMSE).

Furthermore, we implement the Gaussian process (GP) with a sliding window scheme to predict online drone trajectories. Recognizing the high computational complexity of the GP, we also introduce a low-rank approximation method, structured kernel interpolation (SKI) GP, aiming to conserve computational resources. Finally, we compare the prediction performances of the IMM filter, classical GP, and SKI GP on real drone trajectories. The results highlight that the classical GP method enhanced prediction accuracy, achieving an RMSE of less than 1.7m, which is 50% lower compared to the model-based IMM filter. Additionally, the SKI GP realizes a 25% reduction in computation time compared to the classical GP, despite a slight compromise in prediction accuracy.

As unmanned aerial systems (UAS) turn into a full-fledged industry, the sky will be much more crowded in the future. Large-scale UAV applications make reliable UAV navigation a pressing need. Traditionally, global navigation satellite system (GNSS) is extensively used as the primary positioning, navigation, and timing (PNT) service. However, GNSS is vulnerable to intentional radio interference such as spoofing, jamming, and repeating. Hence, alternative PNT (APNT) attracted many researchers' attention.

In this thesis, instead of GNSS signals, ADS-B signals from piloted aircraft are leveraged for UAV navigation. We propose a cooperative navigation strategy for multiple UAVs in GNSS-denied environments. It consists of: 1) a system-level, leader-follower cooperative strategy; 2) a sensor fusion algorithm for individual UAV navigation based on the extended Kalman filter. Furthermore, the effects of asynchronous clocks are studied and a joint relative positioning and synchronization algorithm is applied to tackle this problem.

Finally, Monte Carlo experiments in a multi-UAV scene are performed to verify the proposed algorithms. The results show that the proposed algorithms achieve a performance comparable to civilian GNSS on the selected data set and under the system assumptions we made. Moreover, the proposed cooperative navigation framework only needs one ground station of limited service capacity as external aid. Compared with large-scale, specialized terrestrial APNT service networks, our proposed framework is more flexible and the system can be deployed in areas without infrastructure.

This work is focused on the distributed system, i.e. Multi-agent Systems (MAS), with application in environmental monitoring and learning. The specific task is to develop algorithms, i.e. Gaussian Process (GP), that are robust, accurate and fully-distributed to learn the unknown spatial environmental field. The two main problems are (1). how to optimize GP hyperparameters in fully-distributed manner, and (2). how to aggregate predictions from agents.

The state-of-the-art solution for distributed GP hyperparameter optimization problem is proximal alternated direction method of multipliers (pxADMM) algorithm, which requires a center station in MAS. Based on pxADMM, two fully-distributed pxADMM algorithms are proposed such that the center station is no longer needed. Asynchronous behavior is also introduced into the proposed algorithms, so that they can deal with heterogeneous processing time of agents. Simulations are carried out on both artificial and real datasets. Results show that the proposed methods all achieve stable convergence.

The aggregation methods can be classified based on whether the local datasets are assumed to be independent or not. Under independent assumption, PoE and BCM families of methods can be distributed by applying discrete time consensus filter (DTCF). In this project, primal-dual method of multiplier (PDMM) is proposed to replace DTCF so that the aggregation converges faster. Without independent assumption, the Nested Pointwise Aggregation of Experts (NPAE) considers the cross-correlation among local datasets to achieve consistent aggregation. The current NPAE-JOR algorithm distributes NPAE in complete graph, where flooding variables across network is required before aggregation. In this project, CON-NPAE is proposed to extend NPAE to fully-distributed version in connected graph, where flooding is not required. Simulations on artificial and real datasets are performed. Results show that the proposed PDMM based algorithm reduces the iterations needed for fully-distributed PoE and BCM families of methods. The CON-NPAE is fully-distributed and makes better aggregations than independence assumption based methods in networks with high connectivity. The connection between its performance and MAS structure requires further study.

In conclusion, fully-distributed and asynchronous algorithms are proposed for GP hyperparameter optimization based on pxADMM. The fully-distributed PoE and BCM methods are accelerated by applying PDMM. The CON-NPAE is proposed to make NPAE fully-distributed.

In future work, the theoretical convergence of fully-distributed pxADMM should be researched. For GP aggregation, the effect of network structure on the performance of CON-NPAE can be studied. Also, inducing points method is a possible solution to alleviate the flooding overhead of distributed NPAE.

Distributed formation control has received increasing attention in multiagent systems. Maintaining certain geometry in space is advantageous in many applications such as space interferometry and underwater sensing. At present, there is a variety of distributed solutions for agents to converge to desired formations and track a series of prescribed maneuvers. They typically rely on the relative kinematics e.g., relative positions of the neighboring agents as state observations for the local controller. In harsh working environments, the acquisition of the relative kinematics is challenged and observation losses might occur, which can be detrimental to the optimality of formation.

In this work, observation losses in noisy environments are addressed under a distributed formation control framework. Three types of solutions are proposed to enhance the robustness which is evaluated through the improvements of tracking error, convergence speed, and smoothness of trajectories in both random and permanent loss settings. Firstly, a relative localization technique is proposed using formation itself as a spatial constraint. Secondly, a dynamic model is established for the agents entailed by a Kalman filter-based solution. Finally, a fusion of the previous two types is inspired and it exhibits superior performance than both aforementioned types individually.

This work not only provides means of relative localization without additional sensor data but also shares insights into coping with random or permanent graph changes for stress-based formation control systems. This could potentially lead to the exploration of formation control with subgraphs or energy-efficient sensing as future directions.

Autonomous agents are the future of many services and industries such as delivery systems, surveillance and monitoring, and search and rescue missions. An important aspect in an autonomous agent is the navigation system it uses to traverse the environment. Not much emphasis has been paid in the past on autonomous agent navigation in cluttered environments. Cluttered and unknown environments such as forests and subaquatic environments need to have autonomous navigation systems developed just for them due to their uncertain and changing nature. Path planning algorithms are used for the navigation of an autonomous agent in an environment. The agent needs to reach a target location while avoiding the obstacles it detects along the path. Such a system is called a Detect and Avoid (DAA) system and there are different implementations for it of which some are explored in this thesis. The Artificial Potential Fields method or APF for short is a method for mobile agent navigation which is based on generating an attractive force on the agent from the target and a repulsive force from the obstacles. This leads to the agent reaching the target while avoiding the obstacles along the way. The Classical APF (CAPF) method works for structured environments well but not for cluttered environments. The CAPF method can be replaced with a modified version where the agent is surrounded by a set of points (called bacteria points) around its current location and the agent moves by selecting a bacteria point as a future location. This method is named the Bacteria APF (BAPF) method. This selection happens through combinatorial optimization based on the potential value of each bacteria point. In this thesis, we propose two distinct contributions to the BAPF method. The first one being the use of an adaptive parameter in the repulsive cost function which is determined through a brute-force search. The second addition is a branching cost function that changes the value of the repulsive potential based on predefined perimeters around each obstacle. We show through simulations on densely and lightly cluttered environments that this Improved BAPF (IBAPF) method significantly improves the performance of the system in terms of the convergence to the target by almost 200% and reduced the time it takes to converge by around 25% as well as maintain the safety of the navigation route by keeping the average distance from obstacles around the same value.

The recent advances in wireless communication, micro-fabrication, and integration have led to the rise of multi-agent networks such as wireless sensor networks, drone swarms, and satellite arrays. These multi-agent networks comprise multiple agents, which cooperate to solve a problem collectively, beyond the individual knowledge of a single agent. Multi-agent networks are also known for improving overall system performance, especially in terms of scalability, robustness, and flexibility.

Multi-agent networks are tasked with a variety of tasks e.g., target tracking, surveillance, traffic control, and environmental monitoring. Traditionally these tasks are solved using distributed state-space filtering methods, however under the assumptions of linear models with underlying Gaussian noise. However, in general, the models could be Non-linear, and the underlying stochastic noise is not necessarily Gaussian in nature. To address this limitation, Distributed particle filters (DPF) have been investigated in recent years. Depending on the quantities exchanged in the network, DPFs can be categorized as weight-based, likelihood-based, or posterior-based.

One of the key challenges for effective implementation of the DPF is the need for approximation strategies to reduce the burden on communication and computation resources to suit engineering applications. To this end, we review two state-of-the-art DPF, (a) one based on applying the Graph Laplacian (GL) approximation to weights, and (b) the other using the Gaussian mixture model (GMM) to represent the local posteriors. These two algorithms significantly reduce the communication burden; however, the main limitation is the synchronization and heavy computational burden, which places high demands on hardware and software implementation of agents.

To improve on these existing methods, we propose the use of Gaussian process (GP) enhanced resampling algorithm, which reduces the particle set size, while ensuring an acceptable filter performance. Based on this GP-enhanced resampling scheme, we propose a novel Distributed particle filtering algorithm. Secondly, based on the proposed GP-enhanced sampling, we propose several metaheuristic optimization algorithms (i.e., genetic algorithm and firefly algorithm) to fuse the information across the whole network to improve the estimation performance (i.e., to seek the global optimal particle set as we can treat particles as “parents” or “fireflies”).

We compare our proposed algorithms and study their performance against the state-of-the-art DPFs in terms of time, space, and communication complexity, for a target tracking application. We use time-averaged root mean square error (ARMSE), which is used to quantify the performance of each DPF. Our numerical results demonstrate the superiority of our proposed algorithms under the condition of limited communication and computational resources, and still maintaining the optimality of particle filter.

For example, to achieve a certain ARMSE for a target tracking application, the traffic for existing GL and GMM based DPFs, is 12000 scalars/second/agent and 600 scalars/second/agent respectively, while our proposed GP-enhanced resampling and firefly algorithm uses only 900scalars/second/agent. In terms of the computation runtime, the GL and GMM based approaches take 40 and 300 seconds (about 5 minutes) respectively, while our proposed algorithm takes just 1.5 seconds, and thus shows significant improvement. Finally, we propose research directions to further improve our proposed algorithms in the future.