Multi-robot systems (MRSs) have gained significant momentum as of late in the robotics community as they find application in tasks such as unknown environment exploration, distributed surveillance, and search and rescue. Operating robot teams in real world environments introduces a notion of uncertainty into the system, especially when it comes to the ability of the MRS to reliably communicate. This poses a significant challenge as a stable communication topology is the backbone of the team's ability to coordinate. Additionally, as these systems continue to evolve and integrate further into our society, a growing threat of adversarial attackers pose the risk of compromising nominal operation. As such, this dissertation aims to model the effects of uncertainty in communication on the topology of the MRS using a probabilistic interaction model. More specifically we are interested in studying a probabilistic perspective to those topologies that pertain to the security and resilience of an MRS against adversarial attacks. Having a model that is capable of capturing how probabilistic topologies may evolve over time is essential for secure and resilient planning under communication uncertainty. As a result, we develop probabilistic models, both exact and approximate, for the topological properties of system left-invertibility and (r, s)-robustness that respectively characterize the security and resilience of an MRS. In our modeling, we use binary decision diagrams, convolutional neural networks, matroid theory and more to tackle the problems related to probabilistic security and resilience where we find exact solutions, calculate bounds, solve optimization problems, and compute informative paths for exploration. / Doctor of Philosophy / When robots coordinate and interact together to achieve a collaborative task as a team, we obtain what is known as a multi-robot system or MRS for short. MRSs have several advantages over single robots. These include reliability through redundancy, where several robots can perform a given task in case one of the robots unexpectedly fails. The ability to work faster and more efficiently by working in parallel and at different locations. And taking on more complex tasks that can be too demanding for a single robot to complete. Unfortunately, the advantages of MRSs come at a cost, they are generally harder to coordinate, the action of one robot often depends on the action of other robots in the system, and they are more vulnerable to being attacked or exploited by malicious attackers who want to disrupt nominal operation. As one would expect, communication plays a very important roles in coordinating a team of robots. Unfortunately, robots operating in real world environments are subject to disturbances such as noise, obstacles, and interference that hinders the team's ability to effectively exchange information. In addition to being crucial in coordination, effective information exchange plays a major role in detecting and avoiding adversarial robots. Whenever misinformation is being spread in the team, the best way to counter such adversarial behavior is to communicate with as much well-behaving robots as possible to identity and isolate inconsistencies. In this dissertation we try to study how uncertainty in communication affects a system's ability to detect adversarial behavior, and how we can model such a phenomenon to help us account for these uncertainties when designing secure and resilient multi-robot systems.

Despite the existence of robots that are capable of lifting heavy loads, robotic assistants that can help people move objects as part of a team are not available. This is because of a lack of critical intelligence that results in inefficient and ineffective performance of these robots. This work makes progress towards improved intelligence of robotic lifting assistants by studying human-human teams in order to understand basic principles of co-manipulation teamwork. The effect of modus, or the manner in which a team moves an object together, is the primary study of this work. Data was collected from over 30 human-human trials in which participants in teams of two co-manipulated an object that weighed 60 pounds. These participants maneuvered through a series of five obstacles while carrying the object, exhibiting one of four modi at any given time. The raw data from these experiments was cleaned and distilled into a pose trajectory, velocity trajectory, acceleration trajectory, and interaction wrench trajectory. Classifying on the original base set of four modi with a neural net showed that two of the three modi were very similar, such that classification between three modi was more appropriate. The three modi used in classification were \emph{quickly}, \emph{smoothly} and \emph{avoiding obstacles}. Using a convolutional neural net, three modi were able to be classified from a validation set with up to 85\% accuracy. Detecting modus has the potential to greatly improve human-robot co-manipulation by providing a means to determine an appropriate robot behavior objective function. Survey data showed that participants trust each other more after working together and that they feel that their partners are more qualified after they worked together. A number of modified scales were also shown to be reliable which will allow future researchers in human-robot co-manipulation to properly evaluate how humans feel about working with each other. These same scales will also provide a useful comparison to human-robot teams in order to determine how much humans trust robots as co-manipulation team members.

Using teams of autonomous, heterogeneous robots to operate in dangerous environments has a number of advantages. Among these are cost-effectiveness and the ability to spread out skills among team members. The nature of operating in dangerous domains means that the risk of loss is higher---teams will often lose members and must acquire new ones. In this work, I explore various recruitment strategies for the purpose of improving an existing framework for team management. My additions allow robots to more actively acquire new teams members and assign tasks among other robots on a team without the intervention of a team leader. I evaluate this framework in simulated post-disaster environments where the risk of robot loss is high and communications are often unreliable. My results show that in many scenarios, active recruitment strategies provide significant performance benefits. / February 2017

The use of teams of coordinated mobile robots in industrial settings such as underground mining, toxic waste cleanup and material storage and handling, is a viable and reliable approach to solving such problems that require or involve automation. In this thesis, abilities a team of mobile robots should demonstrate in order to successfully perform a mission in industrial settings are identified as a set of functional components. These components are related to navigation and obstacle avoidance, localization, task achieving behaviors and mission planning. The thesis focuses on designing and developing functional components applicable to diverse missions involving teams of mobile robots; in detail, the following are presented: 1. A navigation and obstacle avoidance technique to safely navigate the robot in an unknown environment. The technique relies on information retrieved by the robot's vision system and sonar sensors to identify and avoid surrounding obstacles. 2. A localization method based on Kalman filtering and Fuzzy logic to estimate the robot's position. The method uses information derived by multiple robot sensors such as vision system, odometer, laser range finder, GPS and IMU. 3. A target tracking and collision avoidance technique based on information derived by a vision system and a laser range finder. The technique is applicable in scenarios where an intruder is identified in the patrolling area. 4. A limited lookahead control methodology responsible for mission planning. The methodology is based on supervisory control theory and it is responsible for task allocation between the robots of the team. The control methodology considers situations where a robot may fail during operation. The performance of each functional component has been verified through extensive experimentation in indoor and outdoor environments. As a case study, a warehouse patrolling application is considered to demonstrate the effectiveness of the mission planning component.

Teams of mobile robots have numerous applications, such as space exploration, underground mining, warehousing, and building security. Multi-robot teams can provide a number of practical benefits in such applications, including simultaneous presence in multiple locations, improved system performance, and greater robustness and redundancy compared to individual robots. This thesis addresses three aspects of coordination and navigation for teams of mobile robots: localization, the estimation of the position of each robot in the environment; motion planning, the process of finding collision-free trajectories through the environment; and task allocation, the selection of appropriate goals to be assigned to each robot. Each of these topics are investigated in the context of many robots working in a common environment. A particle-filter based system for cooperative global localization is presented. The system combines the sensor data from three robots, including measurements of the distances between robots, to cooperatively estimate the global position of each robot in the environment. The method is developed for a single triad of robots, then extended to larger groups of robots. The algorithm is demonstrated in a simulation of robots equipped with only simple range sensors, and is shown to successfully achieve global localization of robots that are unable to localize using only their own local sensor data. Motion planning is investigated for large teams of robots operating in tunnel and corridor environments, where coordinated planning is often required to avoid collision or deadlock conditions. A complete and scalable motion planning algorithm is presented and evaluated in simulation with up to 150 robots. In contrast to popular decoupled approaches to motion planning (which cannot guarantee a solution), this algorithm uses a multi-phase approach to create and maintain obstacle-free paths through a graph representation of the environment. The resulting plan is a set of collision-free trajectories, guaranteeing that every robot will reach its goal. The problem of task allocation is considered in the same type of tunnel and corridor environments, where tasks are defined as locations in the environment that must be visited by one of the robots in the team. To find efficient solutions to the task allocation problem, an optimization approach is used to generate potential task assignments, and select the best solution. The multi-phase motion planner is applied within this system as an efficient method of evaluating potential task assignments for many robots in a large environment. The algorithm is evaluated in simulations with up to 20 robots in a map of large underground mine. A real-world implementation of 3 physical robots was used to demonstrate the implementation of the multi-phase motion planning and task allocation systems. A centralized motion planning and task allocation system was developed, incorporating localization and time-dependent trajectory tracking on the robot processors, enabling cooperative navigation in a shared hallway environment.

Teams of mobile robots have numerous applications, such as space exploration, underground mining, warehousing, and building security. Multi-robot teams can provide a number of practical benefits in such applications, including simultaneous presence in multiple locations, improved system performance, and greater robustness and redundancy compared to individual robots. This thesis addresses three aspects of coordination and navigation for teams of mobile robots: localization, the estimation of the position of each robot in the environment; motion planning, the process of finding collision-free trajectories through the environment; and task allocation, the selection of appropriate goals to be assigned to each robot. Each of these topics are investigated in the context of many robots working in a common environment. A particle-filter based system for cooperative global localization is presented. The system combines the sensor data from three robots, including measurements of the distances between robots, to cooperatively estimate the global position of each robot in the environment. The method is developed for a single triad of robots, then extended to larger groups of robots. The algorithm is demonstrated in a simulation of robots equipped with only simple range sensors, and is shown to successfully achieve global localization of robots that are unable to localize using only their own local sensor data. Motion planning is investigated for large teams of robots operating in tunnel and corridor environments, where coordinated planning is often required to avoid collision or deadlock conditions. A complete and scalable motion planning algorithm is presented and evaluated in simulation with up to 150 robots. In contrast to popular decoupled approaches to motion planning (which cannot guarantee a solution), this algorithm uses a multi-phase approach to create and maintain obstacle-free paths through a graph representation of the environment. The resulting plan is a set of collision-free trajectories, guaranteeing that every robot will reach its goal. The problem of task allocation is considered in the same type of tunnel and corridor environments, where tasks are defined as locations in the environment that must be visited by one of the robots in the team. To find efficient solutions to the task allocation problem, an optimization approach is used to generate potential task assignments, and select the best solution. The multi-phase motion planner is applied within this system as an efficient method of evaluating potential task assignments for many robots in a large environment. The algorithm is evaluated in simulations with up to 20 robots in a map of large underground mine. A real-world implementation of 3 physical robots was used to demonstrate the implementation of the multi-phase motion planning and task allocation systems. A centralized motion planning and task allocation system was developed, incorporating localization and time-dependent trajectory tracking on the robot processors, enabling cooperative navigation in a shared hallway environment.

As the applications of mobile robotics evolve it has become increasingly less practical for researchers to design custom hardware and control systems for each problem. This research presents a new approach to control system design that looks beyond end-of-lifecycle performance and considers control system structure, flexibility, and extensibility. Toward these ends the Control ad libitum philosophy is proposed, stating that to make significant progress in the real-world application of mobile robot teams the control system must be structured such that teams can be formed in real-time from diverse components. The Control ad libitum philosophy was applied to the design of the HAA (Host, Avatar, Agent) architecture: a modular hierarchical framework built with provably correct distributed algorithms. A control system for exploration and mapping, search and deploy, and foraging was developed to evaluate the architecture in three sets of hardware-in-the-loop experiments. First, the basic functionality of the HAA architecture was studied, specifically the ability to: a) dynamically form the control system, b) dynamically form the robot team, c) dynamically form the processing network, and d) handle heterogeneous teams. Secondly, the real-time performance of the distributed algorithms was tested, and proved effective for the moderate sized systems tested. Furthermore, the distributed Just-in-time Cooperative Simultaneous Localization and Mapping (JC-SLAM) algorithm demonstrated accuracy equal to or better than traditional approaches in resource starved scenarios, while reducing exploration time significantly. The JC-SLAM strategies are also suitable for integration into many existing particle filter SLAM approaches, complementing their unique optimizations. Thirdly, the control system was subjected to concurrent software and hardware failures in a series of increasingly complex experiments. Even with unrealistically high rates of failure the control system was able to successfully complete its tasks. The HAA implementation designed following the Control ad libitum philosophy proved to be capable of dynamic team formation and extremely robust against both hardware and software failure; and, due to the modularity of the system there is significant potential for reuse of assets and future extensibility. One future goal is to make the source code publically available and establish a forum for the development and exchange of new agents.

As the applications of mobile robotics evolve it has become increasingly less practical for researchers to design custom hardware and control systems for each problem. This research presents a new approach to control system design that looks beyond end-of-lifecycle performance and considers control system structure, flexibility, and extensibility. Toward these ends the Control ad libitum philosophy is proposed, stating that to make significant progress in the real-world application of mobile robot teams the control system must be structured such that teams can be formed in real-time from diverse components. The Control ad libitum philosophy was applied to the design of the HAA (Host, Avatar, Agent) architecture: a modular hierarchical framework built with provably correct distributed algorithms. A control system for exploration and mapping, search and deploy, and foraging was developed to evaluate the architecture in three sets of hardware-in-the-loop experiments. First, the basic functionality of the HAA architecture was studied, specifically the ability to: a) dynamically form the control system, b) dynamically form the robot team, c) dynamically form the processing network, and d) handle heterogeneous teams. Secondly, the real-time performance of the distributed algorithms was tested, and proved effective for the moderate sized systems tested. Furthermore, the distributed Just-in-time Cooperative Simultaneous Localization and Mapping (JC-SLAM) algorithm demonstrated accuracy equal to or better than traditional approaches in resource starved scenarios, while reducing exploration time significantly. The JC-SLAM strategies are also suitable for integration into many existing particle filter SLAM approaches, complementing their unique optimizations. Thirdly, the control system was subjected to concurrent software and hardware failures in a series of increasingly complex experiments. Even with unrealistically high rates of failure the control system was able to successfully complete its tasks. The HAA implementation designed following the Control ad libitum philosophy proved to be capable of dynamic team formation and extremely robust against both hardware and software failure; and, due to the modularity of the system there is significant potential for reuse of assets and future extensibility. One future goal is to make the source code publically available and establish a forum for the development and exchange of new agents.

When humans and remote robots work together on a team, the robots always interact with a human supervisor, even if the interaction is limited to occasional reports. Distracting a human with robotic interactions doesn't pose a problem so long as the inclusion of robots increases the team's overall effectiveness. Unfortunately, increasing the supervisor's cognitive load may decrease the team's sustainable performance to the point where robotic agents are more a liability than an asset. Present approaches resolve this problem with adaptive autonomy, where a robot changes its level of autonomy based on the supervisor's cognitive load. This thesis proposes to augment adaptive autonomy by modeling temporal latency and using this model to optimally select the temporal interval between when a supervisor is informed of a pending change and when the robot makes the change. This enables robotic team members to time their actions in response to the supervisor's cognitive load. The hypothesis is confirmed in a user-study where 26 participants interacted with a simulated search-and-rescue scenario.

Experience forms the basis of learning. It is crucial in the development of human intelligence, and more broadly allows an agent to discover and learn about the world around it. Although experience is fundamental to learning, it is costly and time-consuming to obtain. In order to speed this process up, humans in particular have developed communication abilities so that ideas and knowledge can be shared without requiring first-hand experience. Consider the same need for knowledge sharing among robots. Based on the recent growth of the field, it is reasonable to assume that in the near future there will be a collection of robots learning to perform tasks and gaining their own experiences in the world. In order to speed this learning up, it would be beneficial for the various robots to share their knowledge with each other. In most cases, however, the communication of knowledge among humans relies on the existence of similar sensory and motor capabilities. Robots, on the other hand, widely vary in perceptual and motor apparatus, ranging from simple light sensors to sophisticated laser and vision sensing. This dissertation defines the problem of how heterogeneous robots with widely different capabilities can share experiences gained in the world in order to speed up learning. The work focus specifically on differences in sensing and perception, which can be used both for perceptual categorization tasks as well as determining actions based on environmental features. Motivating the problem, experiments first demonstrate that heterogeneity does indeed pose a problem during the transfer of object models from one robot to another. This is true even when using state of the art object recognition algorithms that use SIFT features, designed to be unique and reproducible. It is then shown that the abstraction of raw sensory data into intermediate categories for multiple object features (such as color, texture, shape, etc.), represented as Gaussian Mixture Models, can alleviate some of these issues and facilitate effective knowledge transfer. Object representation, heterogeneity, and knowledge transfer is framed within Gärdenfors' conceptual spaces, or geometric spaces that utilize similarity measures as the basis of categorization. This representation is used to model object properties (e.g. color or texture) and concepts (object categories and specific objects). A framework is then proposed to allow heterogeneous robots to build models of their differences with respect to the intermediate representation using joint interaction in the environment. Confusion matrices are used to map property pairs between two heterogeneous robots, and an information-theoretic metric is proposed to model information loss when going from one robot's representation to another. We demonstrate that these metrics allow for cognizant failure, where the robots can ascertain if concepts can or cannot be shared, given their respective capabilities. After this period of joint interaction, the learned models are used to facilitate communication and knowledge transfer in a manner that is sensitive to the robots' differences. It is shown that heterogeneous robots are able to learn accurate models of their similarities and difference, and to use these models to transfer learned concepts from one robot to another in order to bootstrap the learning of the receiving robot. In addition, several types of communication tasks are used in the experiments. For example, how can a robot communicate a distinguishing property of an object to help another robot differentiate it from its surroundings? Throughout the dissertation, the claims will be validated through both simulation and real-robot experiments.