Search

The of engineering microscopic collectives such as bacteria, mammalian cells and microrobots has implications from the design of novel biomedical therapies to the development of functional materials. Controlling microagent behaviour is challenging however, due to the limited capabilities of individual agents, lack of straightforward programmability and difficulties in visualisation. To address this, the Dynamic Optical Micro-Environment (DOME) has been developed as a low-cost, modular and open-source device for closed-loop light-based control of microagent systems at both an individual and collective level. The DOME offers an accessible means to study complex multiagent phenomena and implement new behaviours with desired functionalities. This work explores the state-of-the-art regarding light-based microagent control and the current hardware landscape for optical control systems. The DOME is presented, with details of the low-cost fabrication process and characterisation of key specifications. Control over microsystem behaviour using light is demonstrated through the implementation of building blocks towards swarm control in a system of light-responsive Volvox agents. Future steps explored include the engineering of cellular collectives, such as biofilms and migrating tissue, as well as the potential for integration of machine learning techniques for the discovery of de novo swarm behaviours.

In the quest for the exploration and study of the vast and complex oceanic environments and the biological species that inhabit them, a variety of different methodologies and techniques are currently employed. Among them, the ability to localise and monitor underwater acoustic sources by passively observing the signals that they emit was proven to be a valuable asset for the study of both biological entities (e.g. marine mammals) as well as anthropogenic ones (e.g. transportation vessels, off-shore drilling processes etc.). Traditional marine environment monitoring systems employ a single oceanographic research vessel, equipped with all the necessary sensors for the acquisition of data. This method of study of marine environments is expensive, time-consuming and unable to study efficiently the large areas that need to be explored. Therefore, recent research turned towards the employment of low-cost, semi-disposable, distributed sensing nodes for simultaneous data collection over large distances. The large numbers of nodes used in such systems, combined with their simple, low-cost, decentralised nature matches the ethos of another engineering field - that of swarm robotics. Therefore, it has recently been proposed that swarm intelligence algorithms and techniques could be used for the control of the movement of such marine systems. This thesis introduces several novel swarm intelligence algorithms for use for the control of marine distributed sensor networks with passive acoustic capabilities. This is achieved through the modification of a prevalent swarm intelligence algorithm, particle swarm optimisation (PSO), originally used in numerical optimisation applications. The first half of the thesis is concerned with adapting PSO for the accurate low-level motion control of robotic swarms with the task of source localisation, while also allowing for the inclusion of alternative swarm robotic tasks like obstacle avoidance and aggregation. Afterwards, the thesis focuses on combining PSO with wavefield correlation techniques, currently used in marine passive acoustic systems such as multi-hydrophone arrays. In the end, it is demonstrated how the proposed algorithms can be combined together, enabling a robotic swarm of marine surface vehicles to localise a marine acoustic source, approach and encircle it and continue monitoring it while maintaining a minimum distance from it. All of the presented algorithms are implemented and tested using MATLAB simulations, while some of them are also further validated using Gazebo simulations employing detailed robot models.

The collective robotics task of distributed spatial coverage through random motion is inherently difficult in unbounded environments. In the absence of boundaries, the diffusion of agents is an unavoidable result of stochastic movement that leads to a breaking up of the group. This thesis initiates an investigation of the ways in which the intelligence of individual agents can be used to produce cohesive collective random motion in unbounded environments. A perceptual computation framework is developed. This reveals that when agents are able to model and respond to the system which they are part of, increasingly sophisticated collective behaviour is produced. This framework is used to develop a swarm robotics sociogenesis solution, taking inspiration from the spatial sorting of animal groups according to social dominance hierarchies. Agents are co-ordinated in space using a self-organised artificial social structure, by integrating distributed machine learning methods into the swarm intelligence paradigm. Each agent learns a local regression model of the artificial social structure, and moves with the goal of minimising the difference between its own state and that predicted by its internal model. A one-dimensional sociogenesis movement model is developed and tested in simulation, and achieves self-organised and cohesive spatial coverage. Following this, the model is extended to two-dimensional space and achieves excellent spatial sorting that remains stable up to intermediate simulation times. The final part of this thesis investigates how the motion of single agents becomes bounded through a memory of past occupied states. The influence of memory is formally studied using a generalised Langevin equation (GLE). A set of Fokker-Planck equations is derived to describe the macroscopic properties of the GLE, extending previously derived results. The exact conditions for which the memory of the agent produces non-Markov dynamics are then derived, using the conditional probability of the process. The work presented in chapter 2 is being prepared for submission in Collective Intelligence: Z. Neu and L. Giuggioli, Towards a statistical physics of interacting intelligent agents, 2022. The results of chapter 3 are being prepared for submission in the journal paper Physical Review E: Z. Neu and L. Giuggioli, A sociogenesis movement model for cohesive collective motion in unbounded domains, 2022, and have also resulted in a research grant from Amazon Research Awards to develop the project Multi-robot online exploration in extreme unbounded environments through adaptive socio-spatial ordering, 2022-2023.

Swarm robotics, inspired by swarms in nature, has great potential. The resilience, scalability, robustness and redundancy of having many robots collectively perform tasks such as mapping, disaster recovery, pollution control, and cleaning make for a compelling vision. To achieve this, we need to design swarm robot systems to have a desired collective behaviour, but the design of controllers for the individual robots of a swarm such that this behaviour emerges from the interaction of the individual robots is difficult. Current solutions often use off-line automatic discovery by artificial evolution of robot controllers, which are then transferred into the swarm. This is problematic for two important reasons. Firstly, since there is the need for additional supporting infrastructure, both to evolve the new controllers and to communicate them to the swarm, the swarm is not self-sufficient. Secondly, the evolved controllers are often opaque and hard to understand, an important consideration for safety and explainability reasons. In this work we tackle both of these issues. We build a swarm of robots with very high computing performance using recently available mobile computation devices. This high performance allows us to move the evolutionary process, dependent on processing power for simulation, into the swarm. Because the computational power of the swarm grows with the size of the swarm, it is both autonomous and scalable. We use behaviour trees as the individual robot controller architecture. They are modular, hierarchical and human readable. By developing automatic tools to simplify large evolved trees, we can understand, explain, and even improve the evolved controllers. By moving the evolutionary process into the swarm, and by using understandable controllers, we make the swarm autonomous, scalable, and understandable, necessary steps towards their real-world deployment.

Swarm robotics is the study of developing and controlling large groups of robots. Collectives of robots possess advantages over single robots such as being robust to mission failures due to single-robot errors. Experimental research in swarm robotics is currently limited by swarm robotic technology. Current swarm robotic systems are either small groups of sophisticated robots or large groups of simple robots due to manufacturing overhead, functionality-cost dependencies, and their need to avoid collisions, amongst others. It is therefore useful to develop a swarm robotic system that is easy to manufacture, that utilises its sensors beyond standard usage, and that allows for physical interactions. In this work, I introduce a new type of low-friction locomotion and show its first implementation in the HoverBot system. The HoverBot system consists of an air-levitation and magnet table, and a HoverBot agent. HoverBots are levitating circuit boards which are equipped with an array of planar coils and a Hall-effect sensor. HoverBot uses its coils to pull itself towards magnetic anchors that are embedded into a levitation table. These robots consist of a Printed Circuit Board (PCB), surface mount components, and a battery. HoverBots are easily manufacturable, robots can be ordered populated; the assembly consists of plugging in a battery to a robot. I demonstrate how HoverBot's low-cost hardware can be used beyond its standard functionality. HoverBot's magnetic field readouts from its Hall-effect sensor can be associated with successful movement, robot rotation and collision measurands. I build a time series classifier based on these magnetic field readouts, I modify and apply signal processing techniques to enable the online classification of the time-variant magnetic field measurements on HoverBot's low-cost microcontroller. This method allows HoverBot to detect rotations, successful movements, and collisions by utilising readouts from its single Hall-effect sensor. I discuss how this classification method could be applied to other sensors and demonstrate how HoverBots can utilise their classifier to create an occupancy grid map. HoverBots use their multi-functional sensing capabilities to determine whether they moved successfully or collided with a static object to map their environment. HoverBots execute an "explore-and-return-to-nest" strategy to deal with their sensor and locomotion noise. Each robot is assigned to a nest (landmark); robots leave their nests, move n steps, return and share their observations. Over time, a group of four HoverBots collectively builds a probabilistic belief over its environment. In summary, I build manufacturable swarm robots that detect collisions through a time series classifier and map their environment by colliding with their surroundings. My work on swarm robotic technology pushes swarm robotics research towards studies on collision-dependent behaviours, a research niche that has been barely studied. Collision events occur more often in dense areas and/or large groups, circumstances that swarm robots experience. Large groups of robots with collision-dependent behaviours could become a research tool to help invent and test novel distributed algorithms, to understand the dependencies between local to global (emergent) behaviours and more generally the science of complex systems. Such studies could become tremendously useful for the execution of large-scale swarm applications such as the search and rescue of survivors after a natural disaster.

The ultimate objective of artificial intelligence is to develop intelligent agents that can think and act rationally. In intelligent systems, agents rarely exist in isolation, but instead form part of a larger group of agents all sharing the same (or similar) goals. As such, a population of agents needs to be able to reach an agreement about the state of the world efficiently and accurately, and in a distributed manner, so that they can then make collective decisions. In this thesis we attempt to exploit vagueness in natural language so as to allow agents to be more effective in forming consensus. In classical logic, a proposition can be either true or false, which inevitably leads to situations in which agents that disagree about the truth of a proposition cannot resolve their inconsistencies in an intuitive manner. By adopting an intermediate truth state in cases where there is direct conflict between the beliefs of agents (i.e., where one believes the proposition to be true, and the other believes it to be false), we can combine the beliefs of agents in order to form consensus. We can then repeat this process across the population by forming consensus between agents in an iterative manner, until the population converges to a single, shared belief. This forms the basis of our initial model. We then extend this model of consensus for vague beliefs to take account of epistemic uncertainty. After demonstrating strong convergence properties of both models, we apply our work to a swarm of 400 Kilobot robots,and study the resulting convergence in such a setting. Finally, we propose a model of consensus in which agents attempt to reach an agreement about a set of compound sentences, rather than just a set of propositional variables.

This thesis investigates several aspects of environment-driven adaptation in simulated evolutionary swarm robotics. It is centred around a specific algorithm for distributed embodied evolution called mEDEA. Firstly, mEDEA is extended with an explicit relative fitness measure while still maintaining the distributed nature of the algorithm. Two ways of using the relative fitness are investigated: influencing the spreading of genomes and performing an explicit genome selection. Both methods lead to an improvement in the swarm's abilityto maintain energy over longer periods. Secondly, a communication energy model is derived and introduced into the simulator to investigate the influence of accounting for the costs of communication in the distributed evolutionary algorithm where communication is a key component. Thirdly, a method is introduced that relates environmental conditions to a measure of the swarm's behaviour in a 3-dimensional map to study the environment's influence on the emergence of behaviours at the individual and swarm level. Interesting regions for further experimentation are identified in which algorithm specific characteristics show effect and can be explored. Finally, a novel individual learning method is developed and used to investigate how the most effective balance between evolutionary and lifetime-adaptation mechanisms is influenced by aspects of the environment a swarm operates in. The results show a clearlink between the effectiveness of different adaptation mechanisms and environmental conditions, specifically the rate of change and the availability of learning opportunities.

The field of swarm robotics has been growing fast over the last few years. Using a swarm of simple and cheap robots has advantages in various tasks. Apart from performance gains on tasks that allow for parallel execution, simple robots can also be smaller, enabling them to reach areas that can not be accessed by a larger, more complex robot. Their ability to cooperate means they can execute complex tasks while offering self-organised adaptation to changing environments and robustness due to redundancy. In order to keep individual robots simple, a control algorithm has to keep expensive communication to a minimum and has to be able to act on little information to keep the amount of sensors down. The number of sensors and actuators can be reduced even more when necessary capabilities are spread out over different agents that then combine them by cooperating. Self-organised differentiation within these heterogeneous groups has to take the individual abilities of agents into account to improve group performance. In this thesis it is shown that a homogeneous group of versatile agents can not be easily replaced by a heterogeneous group, by separating the abilities of the versatile agents into several specialists. It is shown that no composition of those specialists produces the same outcome as a homogeneous group on a clustering task. In the second part of this work, an adaptation mechanism for a group of foragers introduced by Labella et al. (2004) is analysed in more detail. It does not require communication and needs only the information on individual success or failure. The algorithm leads to self-organised regulation of group activity depending on object availability in the environment by adjusting resting times in a base. A possible variation of this algorithm is introduced which replaces the probabilistic mechanism with which agents determine to leave the base. It is demonstrated that a direct calculation of the resting times does not lead to differences in terms of differentiation and speed of adaptation. After investigating effects of different parameters on the system, it is shown that there is no efficiency increase in static environments with constant object density when using a homogeneous group of agents. Efficiency gains can nevertheless be achieved in dynamic environments. The algorithm was also reported to lead to higher activity of agents which have higher performance. It is shown that this leads to efficiency gains in heterogeneous groups in static and dynamic environments.

In the field of robotics, multi-agent systems and swarm robotics are emerging topics. As the field continues to grow, swarms of robots are being utilized in various applications and at an ever-growing scale. One of the most commonly referenced works in swarm algorithms is the Boids Algorithm. While Boids is an excellent learning tool for the topic of swarm algorithms and sees applications in computer animation, it does not see direct use in robotics due to the lack of goal-seeking and general object avoidance. We propose a new algorithm to mitigate these limitations by adding goal-seeking and general collision avoidance through the novel use of “ghost boids” without modifying the three base Boids rules or adding significant additions to the control algorithm. Through simulation, we show significant improvements in agent safety and the ability of the algorithm to be successful in a search and rescue task. We also implement a control barrier function on top of our algorithm to make further improvements in safety. We analyze to what extent our algorithms are successful in increasing safety and the ability to complete a mission as well as how the computation time increases as we add more complexity to our algorithms.

Collectives in nature demonstrate behaviors that extend far beyond the capabilities of any single agent. Social slime mold, for example, has thousands of cells that aggregate and form mobile and immobile nutrient-searching structures as a function of the chemical signaling between cells and their interactions with the surrounding environment. This species embodies many features that swarm roboticists wish to incorporate in self-reconfigurable robot collectives: emergent complexity, plasticity, and scalable constituents. This thesis argues that regardless of the length scale, we can implement some of the same principles and features to exploit robot morphology, physical interactions among agents, and low-level coordination mechanisms to enable diverse collective behaviors for useful applications across fields. In this thesis, I present novel emergent behaviors through physical robot collectives at the macro- and micron length scales and virtual collectives and explain how each behavior arises as a function of agents interacting with other agents, agents interacting with or reacting to their environment, and agents exploiting their environment to better influence other agents. I expand the relatively new field of mobile coupled oscillators. Specifically, I present a new iteration of the swarming coupled oscillator (swarmalator) model, which enables many new emergent collective behaviors in addition to those already shown in the literature. It also replicates many behaviors of natural and artificial collectives even though the detailed physical interactions are not taken into account. This abstract model has the potential to inform on the coordination mechanisms possible at different length scales. At the macro-scale communication-constrained coordination may serve as a fall-back, and at smaller length scales it may reveal some collective behaviors that are possible with agents that cannot currently have on-board processing. At the macro-scale, I present soft modular robot collectives that exhibit embodied intelligence by exploiting a tight coupling between sensing and actuation to enable reconfiguration, locomotion, and strain-based consensus among other collective behaviors. I have contributed to the development of various soft actuators, a sensor, and two different types of soft robot collectives with differing capabilities, and have studied these collectives using both physical hardware and simulations. At the micron scale, I present novel behaviors through magnetic microrobot collectives that use physical interactions between agents and their surrounding environment to enable the collective to perform on-demand reconfiguration of various functions and morphologies and manipulate passive objects in the surrounding environment. The microrobot collective behaviors are further studied through a physical model and a swarmalator model. The diverse collective behaviors presented in this thesis hold many potential applications in swarm robotics, self-reconfigurable modular robots, and small-scale manufacturing; additionally, they hold great value for fundamental scientific research in self-assembly, self-organization, and emergent complexity. This thesis expands the fields of soft robotics, microrobotics, and coupled oscillators and demonstrates the potential to unlock many new emergent collective behaviors by exploiting low-level sensing, actuation, and coordination mechanisms.

Swarm robotics is an active research area where large groups of simple robots are developedto perform complex tasks. It has various potential applications such as surveillance, warehouse logistics, and collective construction, because of its large sensing capability, fault tolerance, and collaboration skills that are impossible to a single robot. However, controlling swarm robots to perform high-level tasks automatically and verifiably remains challenging due to the large state space. In this work, I present the verifiable control framework for swarm robots from high-level specifications in different combinations of 4 dimensions: symbolic/continuous levels, centralized/decentralized methods, navigation/formation-based specifications, and non-reactive/reactive tasks. The first contribution of the work is a top-down control framework for non-reactive navigation swarm tasks. At the symbolic level, I propose an integer programming-based method for executing a centralized symbolic plan, and compare it with existing decentralized execution in terms of task efficiency, failure resilience, and computational complexity. At the continuous level, I propose a control barrier function based method which guarantees correct (satisfy all specifications) and safe (no collision) transitions when robots execute the high-level tasks. The trade-offs between centralized and decentralized methods provide guidance to swarm controls to achieve different purposes under different restrictions. The second contribution is a novel abstraction and grammar for location and formation-based swarm specifications. With the proposed abstractions, we developed a centralized control synthesis approach which guarantees that the specified swarm behaviors will be satisfied if feasible. This contribution expands the task space of swarms, and gives insights into controlling a large fleet of autonomous robots to perform complex tasks which require composition of behaviors and coordination of different sub-swarms. The third contribution is an extension to the second, which automatically synthesizes controls for swarms to achieve reactive formation tasks in a decentralized manner. This work utilizes an integer programming-based method to calculate constraints on subswarm sizes given the reactive finite state machine, to ensure a priori feasibility of the task. In addition, a decentralized auction-based algorithm is developed for executing the reactive finite state machine, which satisfies the robot number constraints. Such decentralized frameworks increase the scalability of control synthesis and can be applied to swarm systems with a larger number of robots. The last contribution of this thesis is an automatic approach to redistributing robots according to user requests during the task execution while maintaining the original specifications. The approach includes creating symbolic plans, limiting robot numbers given the constraints, and ensuring all robots to satisfy safety requirements of the given highlevel tasks. This work further enhance the flexibility and reactivity of swarms performing high-level tasks.

In this dissertation, we consider the application of pulse-coupled oscillator theory to real-world, physical networks. When the phase of an oscillator is associated with a physical measure, such as clock timing or robotic heading, discontinuous adjustments of the oscillator's phase is undesirable and potentially disadvantageous. Rather, continuous adjustment of the oscillator phase value is needed over a certain amount of time. To ensure that both synchronization and desynchronization can still be achieved under the constraint of continuous phase value changes, we pursue a novel approach to analyze the generalization of a pulse-coupled oscillator network with a time-varying coupling strength. We provide rigorous mathematical proof for both pulse-coupled synchronization and desynchronization under the proposed phase continuity methods. We then correlate the continuous phase change of the oscillator to a specifically time-varying coupling strength of the network. To verify the analysis, we provide both simulated and experimental results for various synchronization and desynchronization algorithms using the proposed phase continuity methods. Additionally, an oscillator may need to adjust its phase response to received pulses from connected neighbors in the network due to non-ideal conditions of physical systems, such as pulse propagation delay, non-identical oscillator frequencies, and general network topologies, once the network has been deployed. Direct analysis of pulse-coupled oscillator networks under non-ideal conditions is difficult, so we consider a novel approach of using reinforcement learning techniques to have the oscillators use their own experience to approximate an optimal phase response. Using appropriate measures to have a single oscillator estimate the state of the rest of the network, we determine a novel phase response function model in terms of the network topology. The optimality of the proposed phase response function is verified with simulated comparisons to existing synchronization algorithms.

Artificial morphogenesis (or morphogenetic engineering) seeks inspiration from developmental biology to engineer self-organizing systems. The Morphgen language uses partial differential equations (PDEs) to express artificial morphogenetic processes as spatial fields describing large numbers of agents in the continuum limit. I present an approach to compile such systems of PDEs by discretizing their behavior to derive controllers for finite numbers of agents of finite size. This approach builds on a generalization of methods to control swarms of robots based on the computational fluid dynamics technique of smoothed particle hydrodynamics (SPH). I address potential scalability and efficiency challenges in SPH robotics by proposing embodied kernel functions that allow key operations to be offloaded to the physical environment. These kernel functions assume the ability of agents to produce local physical fields with certain properties. I especially explore the possibility of cell-like agents able to control the secretion rate of morphogens that diffuse and are degraded in an aqueous medium. I present strategies to mitigate gaps between simplifying assumptions and expected realities. I explore in simulation the validity, accuracy, and robustness of these strategies and of my overall approach.

Multirobot systems can accomplish complex tasks when they can effectively share data. These tasks can range from search and rescue operations to path mapping operations. Currently Robot Operating System (ROS) framework is the widely used platform for creating a multirobot system, but its usability suffers from the complexity of being a framework, making it hard to employ on a variety of systems and projects. In this study, a multirobot system library was developed as an alternative to the ROS framework. The multirobot system development library was able to effectively share an unrestricted range of data and data types with a measured latency lower than a twentieth of that experienced with ROS. The multirobot system development library was also able to automatically pair new devices to its system unlike the ROS framework. When given the task of tracking displacement, the multirobot system development library was able to use its data sharing methods to create graphics in real-time, representing that the robots travel accurately. When given the task of searching for an object, the robot was able to remotely process large datasets allowing it to identify the target without having the necessary computational to do so alone.

Intelligent robots have been widely studied and investigated to replace, fulfilling a complex mission in a hazardous environment. Lately, swarm robotics, a group of collaborative robots, has become popular because it offers benefits over a single intelligent system. Many strategies have been developed to achieve collective and decentralised control applying evolutionary algorithms. However, since the evolutionary algorithm relies principally on an individual fitness function to explore the solution space, achieving swarm robotics' collaborative behaviour in a dynamic environment becomes a problem. This is due to the lack of adaptation in most of the evolutionary methods. In order to thrive in such environment, external stimuli and rewards from the environment should be utilised as ``knowledge'' to achieve the intelligent behaviour currently lacking in evolutionary swarm robotics. The aims of this research are: (1) to develop novel reward-based evolutionary swarm learning using mechanisms of epigenetic inheritance; and (2) to identify an efficient learning method for the epigenetic layer achieving a decision-making strategy in a dynamic environment. This research's contributions are the development of reward-based co-learning algorithm and co-evolution using epigenetic-based knowledge backup. The reward-based co-learning algorithm enables the swarm to obtain knowledge of the dynamic environment and override the objective-based function to evaluate internal and external problems. An advantage of this is that the learning mechanism also enables the swarm to explore potentially better behaviour without the constraint of an ill-defined objective function. Simulated search-and-rescue missions using a swarm of UAVs shows that individual behaviour evolves differently although each member has the same physical characteristics and the same set of actions. As an addition to reward-based multi-agent learning mechanisms, epigenetics is introduced as a decision-making layer. The epigenetic layer has two functions: there are genetic regulators, as well as an epigenetic inheritance (the epigenetic mechanism). The first is the function of an epigenetic layer regulating how genetic information is expressed as agent’s behaviour (the ``phenotype''). Thus, utilising the regulatory function, the agent is able to switch genetic strategy or decision-making based on external stimulus from the aforementioned reward-based learning. The second function is that epigenetic inheritance enables sharing of genetic regulation and decision-making layer between agents. In summary, this research extends the current literature on evolutionary swarm robotics and decentralised multi-agent learning mechanisms. The combination of both advances the decentralised mechanism in obtaining information and improve collective behaviour.

Human-swarm interaction (HSI) is a research area that studies how human and swarm capabilities can be combined to successfully perform tasks that exceed the performance limits of single robot systems. The main objective of this thesis is to improve the success of HSI by improving the effectiveness of three interdependent elements: the swarm decision-making algorithm in performing its tasks, human interventions in swarm operation, and the interface between the human and the swarm. Swarm decision-making warrants investigation as the state-of-the-art algorithms perform very poorly under some conditions. Analysing the root causes of such failures reveals that a key performance inhibitor is the unreliable estimation of swarm members’ confidence in their judgements. Two different approaches are proposed to circumvent the identified issues. Performance evaluation under different conditions demonstrates the merits of the proposed approaches and shows that profound improvements to the effectiveness and efficiency of swarm decision-making are possible through the reliable estimation of confidence. Improving swarm effectiveness begets significant benefits to mission performance, but it can negatively affect the effectiveness of human interventions. Previous research has shown that when interacting with a highly reliable machine, humans tend to over-rely on the machine and exhibit notable complacency that limits their ability to detect and fix machine errors. Although over-trust in automation is widely blamed for such complacency, this attribution is yet to be empirically confirmed. This gap is addressed through an empirical investigation of trust in HSI. The results confirm the significant role of trust as a predictor of human reliance on swarm, which suggests that designing trust-aware HSI systems may reduce the negative impacts of human reliance. Utilising a highly reliable swarm while maintaining human vigilance is an objective that might not be possible without an effective human-swarm interface. As automation transparency has proven useful for boosting human understanding of machine operations, it could facilitate human awareness of machine limitations and possible failures. Thus, the thesis empirically examines the efficacy of swarm transparency as a potential intervention for minimising human complacency. The results assert the benefits of transparency in ensuring continued human contributions to the mission even when a highly reliable swarm is used.

This dissertation addresses fundamental aspects of re-configurable swarms of Unmanned Underwater Vehicles (UUVs) for Underwater Wireless Optical Networks (UWONs) at different scales. At the scale of the swarm, we propose Guidance, Navigation and Control (GNC) systems to enable permanent UWONs. At the scale of the UUV, we present the development of a man-portable hybrid UUV with high-bandwidth wireless communication. Finally, we investigate, at the scale of the Free-Space Optical (FSO) system, Pointing, Acquisition and Tracking (PAT) systems to enable free-space laser communication. This dissertation begins with the report of a demonstration of underwater laser beam relay with two UUVs to illustrate the concept of swarms of UUVs for UWONs. Specifically, two submarine models were modified, localized and controlled to reflect the laser beam stemming from a laser pointer at the bottom of a water tank to the desired location. In the second part of this dissertation, we show that a guidance system combining Decentralized Model Predictive Control with graph theory enables to steer UUVs such that there is continuously a path through the UUV network to relay the optical signal for an unlimited time by re-configuring successive subfleets where a UUV is always within communication range with the neighbouring subfleets. The guidance system forms the high-level block of a hierarchical control scheme with an Adaptive Feedback Linearization Controller (AFLC) and an Extended Kalman Filter (EKF) at the low-level. In the third part of this dissertation, we present a man-portable hybrid UUV equipped with a LED modem on a Pan-Tilt system enabling 3.5Mbps full-duplex wireless communication over a range of 7m in a wave tank despite the motion of the UUV. The hybrid capabilities of the UUV have been simulated with a Software in the Loop (SITL) simulator, which showed the good performance of the AFLC and EKF implemented in the UUV's autopilot. Finally, the fourth part of this dissertation investigates two PAT systems with different configurations. The first configuration enables full PAT capabilities but requires precise alignment and calibration of the opto-mechanical components while a second configuration enables only tracking but has lower manufacturing requirements.

Synchronization behavior universally occurs in our daily life, and it is so important and ubiquitous that has been studied over the past few decades. For example, the Millennium Bridge nicknamed as "Wobbly Bridge" because of the alarming swaying motion caused by synchronization of pedestrians on its opening day. Nowadays, along with the improvement of theoretical research, scientists start to control swarm of robots based on the idea of synchronization. However, no previous study has investigated the synchronization among indirectly coupled robots, which are completely independent with each other without any central controller. This thesis describes the design and implementation of a new kind of swarm of robots and their decentralized controllers. Firstly, robots with one rear driven legs are designed to form swarms of robots. Then, an adaptive control method based on Hopf oscillator is implemented to regulate the phase of the oscillating walking robots. Data from both simulations and experiments has been collected and analysed to discover the properties of collective behavior. Finally, the significance of synchronization of multi-robot working systems is verified to be lower power consumption and greater total output force compared with disordered or non-regulated systems.

Social insects, including army ants (genus Eciton) self-assemble dynamic structures which allow them to perform more efficiently and achieve otherwise impossible feats. Army ant bridges, for example, grow and shrink according to traffic levels because they incorporate smart, flexible materials (i.e. living ants) and emerge from local behavior. Similarly, self-assembled structures that respond to local stimuli could allow robots to adapt to new conditions when exploring unknown or rapidly changing environments. While many robot systems use rigid modules to create pre-determined structures, the work of my thesis, Eciton robotica, takes a new approach: inspired by army ants, soft robots create amorphous structures whose size and shape can adapt to the needs of the group. Eciton robotica incorporates several novel features: the soft body, the gripping mechanism, communication through vibration, and the self-assembly algorithm. The robots’ corkscrew gripper connects and disconnects easily to Velcro loop and supports loads in tension, sheer and peel; it allows E. robotica robots to climb over other robots and attach to them at any point. In simulation, a simplified model of the robot was used to test a local control rule for adaptive self-assembly: robots stop and become part of a structure when stepped on. Using only local rules and information, simulated robots built and dissolved bridges in response to traffic and varying terrain. To implement this in hardware, the robots send vibration pulses as they walk, and sense this vibration if they are stepped on by other robots. Using vibration sensing and the local control rule, I used two E. robotica robots to form and dissolve adaptive structures in a 2D vertical plane. This work includes a number of firsts: The Flippy robot, an early robot module, is one of the first soft climbing robots, and highly capable though limited to Velcro and fabric based surfaces. The E. robotica platform, to my knowledge, is the first use of soft mobile robots for any collective behavior, including self-assembly, and the first case of adaptive self-assembly in a vertical plane.

Searching and collecting multiple resources from large unmapped environments is an important challenge. It is particularly difficult given limited time, a large search area and incomplete data about the environment. This search task is an abstraction of many real-world applications such as search and rescue, hazardous material clean-up, and space exploration. The collective foraging behavior of robot swarms is an effective approach for this task. In our work, individual robots have limited sensing and communication range (like ants), but they are organized and work together to complete foraging tasks collectively. An efficient foraging algorithm coordinates robots to search and collect as many resources as possible in the least amount of time. In the foraging algorithms we study, robots act independently with little or no central control. As the swarm size and arena size increase (e.g., thousands of robots searching over the surface of Mars or ocean), the foraging performance per robot decreases. Generally, larger robot swarms produce more inter-robot collisions, and in swarm robot foraging, larger search arenas result in larger travel distances causing the phenomenon of diminishing returns. The foraging performance per robot (measured as a number of collected resources per unit time) is sublinear with the arena size and the swarm size. Our goal is to design a scale-invariant foraging robot swarm. In other words, the foraging performance per robot should be nearly constant as the arena size and the swarm size increase. We address these problems with the Multiple-Place Foraging Algorithm (MPFA), which uses multiple collection zones distributed throughout the search area. Robots start from randomly assigned home collection zones but always return to the closest collection zones with found resources. We simulate the foraging behavior of robot swarms in the robot simulator ARGoS and employ a Genetic Algorithm (GA) to discover different optimized foraging strategies as swarm sizes and the number of resources is scaled up. In our experiments, the MPFA always produces higher foraging rates, fewer collisions, and lower travel and search time than the Central-Place Foraging Algorithm (CPFA). To make the MPFA more adaptable, we introduce dynamic depots that move to the centroid of recently collected resources, minimizing transport times when resources are clustered in heterogeneous distributions. Finally, we extend the MPFA with a bio-inspired hierarchical branching transportation network. We demonstrate a scale-invariant swarm foraging algorithm that ensures that each robot finds and delivers resources to a central collection zone at the same rate, regardless of the size of the swarm or the search area. Dispersed mobile depots aggregate locally foraged resources and transport them to a central place via a hierarchical branching transportation network. This approach is inspired by ubiquitous fractal branching networks such as animal cardiovascular networks that deliver resources to cells and determine the scale and pace of life. The transportation of resources through the cardiovascular system from the heart to dispersed cells is the inverse problem of transportation of dispersed resources to a central collection zone through the hierarchical branching transportation network in robot swarms. We demonstrate that biological scaling laws predict how quickly robots forage in simulations of up to thousands of robots searching over thousands of square meters. We then use biological scaling predictions to determine the capacity of depot robots in order to overcome scaling constraints and produce scale-invariant robot swarms. We verify the predictions using ARGoS simulations. While simulations are useful for initial evaluations of the viability of algorithms, our ultimate goal is predicting how algorithms will perform when physical robots interact in the unpredictable conditions of environments they are placed in. The CPFA and the Distributed Deterministic Spiral Algorithm (DDSA) are compared in physical robots in a large outdoor arena. The physical experiments change our conclusion about which algorithm has the best performance, emphasizing the importance of systematically comparing the performance of swarm robotic algorithms in the real world. We illustrate the feasibility of implementing the MPFA with transportation networks in physical robot swarms. Full implementation of the MPFA in an outdoor environment is the next step to demonstrate truly scalable and robust foraging robot swarms.

This paper documents the design and subsequent construction of a low-cost, flexible robotic platform for swarm robotics research, and the selection of appropriate swarm algorithms for the implementation of a swarm focused predominantly on target location. The design described herein is intended to allow for the construction of robots large enough to meaningfully interact with their environment while maintaining a low per-robot cost of materials and a low assembly time. The design process is separated into three stages: mechanical design, electrical design, and software design. All major design components are described in detail under the appropriate design section. The BOM for a single robot is also included, along with relevant testing information.

Swarm intelligence systems, often inspired by biological systems, have proven useful in solving a wide range of complex, real-world problems. The emergent nature of swarming behaviour is key to swarm intelligence systems' ability to solve complex problems, but also limits their usefulness in some problem areas because of the unpredictability of their behaviour. Improved objective measures are required to enhance the predictability of swarming behaviour. Objective measures developed to identify chaotic behaviour - which shares many characteristics with swarming behaviour - have been applied in limited cases to swarming behaviour and show promise for predicting swarming behaviour. From a subjective perspective, to date, there has been no structured analysis of the conditions under which humans identify swarming behaviour or whether there are differences between individuals in the identification of swarming behaviour. The first stage of this study uses a reduced version of Reynolds' original boids model to identify 'rule-sets' - i.e. combinations of attract, align and repel forces - that lead to swarming behaviour. Using these rule-case, the second stage of the study investigates human subjective identification of swarming behaviour and tests objective measures, including chaos measures, for suitability to quantify swarming behaviour. The third stage of the study builds on the first and second stages and develops classifier models and point prediction models to quantify and predict swarming behaviour. Application of chaos measures shows that swarming behaviour is also chaotic behaviour. Further, the chaos measures applied indicate there is no binary swarming/not- swarming situation but rather where swarming occurs there are degrees of 'swarminess'. The investigation finds that human subjects identify swarming behaviour where predicted by objective measures and their responses also suggest there are degrees of 'swarminess' with individuals having differing 'swarminess thresholds'. Sex of subject is identified as one factor that affects an individual's 'swarminess threshold'. The study concludes that a proposed composite measure which includes chaos measures can be used to objectively identify, quantify and predict swarming behaviour and predict subjective identification.of swarming behaviour. Chaos measures therefore offer the potential to improve the predictability and the usability of swarm intelligence systems.

Swarm robotics is a field dedicated to the study of the design and development of certain multi-robot systems. Often times, these groups prove to be more beneficial than a single complex robot as swarms typically provide a more robust and potentially more efficient solution. One such case is the task of escorting a specified target while addressing any potential threats discovered in the environment. In this work, a control algorithm for a high volume, decentralized, homogeneous robot swarm was developed based upon a technique commonly used to model incompressible fluids known as Smoothed Particle Hydrodynamics (SPH). This proposed solution to the asset protection problem was tested against a more commonly accepted method for robot navigation known as potential fields. An alternate algorithm was developed based on this technique and manipulated to perform the same basic duty of asset protection. Both algorithms were tested in simulation using ARGoS as an environment and Swarmanoid’s Footbots as robot models. Five experiments were run in order to examine the functionality of both of these algorithms in relation to formation control and the protection of a mobile asset from mobile threats. The results proved the proposed SPH based algorithm comparable to the potential fields based method while minimizing the escape window and having a slightly higher response rate to introduced threats. These results hint that the concept of using fluid models for control of high volume swarms should further be explored and seriously considered as a potential solution to the asset protection problem.

Traditionally robots are controlled using devices like joysticks, keyboards, mice and other similar human computer interface (HCI) devices. Although this approach is effective and practical for some cases, it is restrictive only to healthy individuals without disabilities, and it also requires the user to master the device before its usage. It becomes complicated and non-intuitive when multiple robots need to be controlled simultaneously with these traditional devices, as in the case of Human Swarm Interfaces (HSI). This work presents a novel concept of using human bio-signals to control swarms of robots. With this concept there are two major advantages: Firstly, it gives amputees and people with certain disabilities the ability to control robotic swarms, which has previously not been possible. Secondly, it also gives the user a more intuitive interface to control swarms of robots by using gestures, thoughts, and eye movement. We measure different bio-signals from the human body including Electroencephalography (EEG), Electromyography (EMG), Electrooculography (EOG), using off the shelf products. After minimal signal processing, we then decode the intended control action using machine learning techniques like Hidden Markov Models (HMM) and K-Nearest Neighbors (K-NN). We employ formation controllers based on distance and displacement to control the shape and motion of the robotic swarm. Comparison for ground truth for thoughts and gesture classifications are done, and the resulting pipelines are evaluated with both simulations and hardware experiments with swarms of ground robots and aerial vehicles.

Human beings are driven to explore distant new worlds as we seek to better understand our place in the Universe. Because of the inherent dangers of human spaceflight, we often send robots as surrogate explorers, controlled from millions of miles away by teams of capable rover drivers here on Earth. As technology continues to advance, scientists and engineers aspire to build low-cost, durable, fully autonomous rovers to succeed today's tele-operated extraplanetary explorers. Here we aim to advance this goal by designing and programming robots that can successfully navigate unknown and variable environments. We present a swarm robotics system that mimics the foraging behaviors of seed-harvester ants, employing evolutionary computation and machine learning to mitigate the adverse effects of unreliable information, variable environments, congestion bottlenecks, and sparse resources. We describe a central-place foraging algorithm (CPFA) whose parameters are evolved by a genetic algorithm (GA) to maximize foraging performance under different experimental conditions. We find that foraging for resources in heterogeneous clusters requires more complex communication, memory, and environmental sensing than strategies evolved in previous work. Additionally, we observe sub-linear scaling in resources collected per robot as swarm size increases, which we attribute to the 'bottleneck' constraint imposed by central-place foraging. Finally, we augment our foraging robot swarm with machine learning and statistical models, demonstrating that combining our existing biologically-inspired CPFA with a cluster exploitation algorithm produces more efficient total resource collection compared to each algorithm acting alone. While our system is designed to be a demonstration platform for swarm robotics research, this work provides a foundation for designing and implementing autonomous robot swarms that can function outside of the academic research laboratory. The ability of robot swarms to tolerate sensor noise, adapt to variable environments, distribute work across large teams, and identify and exploit heterogeneously-distributed resources are all critical factors for successful remote exploration missions on distant worlds.

Interest in swarm robotics, particularly those modeled on biological systems, has been increasing with each passing year. We created the iAnt robot as a platform to test how well an ant-inspired robotic swarm could collect resources in an unmapped environment. Although swarm robotics is still a loosely defined field, one of the included hallmarks is multiple robots cooperating to complete a given task. The use of multiple robots means increased cost for research, scaling often linearly with the number of robots. We set out to create a system with the previously described capabilities while lowering the entry cost by building simple, cheap robots able to operate outside of a dedicated lab environment. Obstacle avoidance has long been a necessary component of robot systems. Avoiding collisions is also a difficult problem and has been studied for many years. As part of moving the iAnt further towards the real-world we needed a method of obstacle avoidance. Our hypothesis is that use of biological methods including evolution, stochastic movements and stygmergic trails into the iAnt Central Place Foraging Algorithm (CPFA) could result in robot behaviors suited to navigating obstacle-filled environments. The result is a modification of the CPFA to include pheromone trails, CPFA-Trails or CPFAT. This thesis first demonstrates the low-cost, simple and robust design of the physical iAnt robot. Secondly we will demonstrate the adaptability of the the system to evolve and succeed in an obstacle-laden environment.

The task of searching a space is critical to a wide range of diverse applications such as land mine clearing and planetary exploration. Because applications frequently require searching remote or hazardous locations, and because the task is easily divisible, it is natural to consider the use of multi-robot teams to accomplish the search task. An important topic of research in this area is the division of the task among robot agents. Interrelated with subtask assignment is failure handling, in the sense that, when an agent fails, its part of the task must then be performed by other agents. This thesis describes Hopscotch, a multi-agent search strategy that divides the search area into a grid of lots. Each agent is assigned responsibility to search one lot at a time, and upon completing the search of that lot the agent is assigned a new lot. Assignment occurs in real time using a simple contract net. Because lots that have been previously searched are skipped, the order of search from the point of view of a particular agent is reminiscent of the progression of steps in the playground game of Hopscotch. Decomposition of the search area is a common approach to multi-agent search, and auction-based contract net strategies have appeared in recent literature as a method of task allocation in multi-agent systems. The Hopscotch strategy combines the two, with a strong focus on robust tolerance of agent failures. Contract nets typically divide all known tasks among available resources. In contrast, Hopscotch limits each agent to one assigned lot at a time, so that failure of an agent compels re-allocation of only one lot search task. Furthermore, the contract net is implemented in an unconventional manner that empowers each agent with responsibility for contract management. This novel combination of real-time assignment and decentralized management allows Hopscotch to resiliently cope with agent failures. The Hopscotch strategy was modeled and compared to other multi-agent strate- gies that tackle the search task in a variety of ways. Simulation results show that Hopscotch is failure-tolerant and very effective in comparison to the other approaches in terms of both search time and search efficiency. Although the search task modeled here is a basic one, results from simulations show the promise of using this strategy for more complicated scenarios, and with actual robot agents.