Simultaneous localization and mapping (SLAM) is a challenging and vital problem in robotics. It is important in tasks such as disaster response, deep-sea and cave exploration, in which robots must construct a map of an unknown terrain, and at the same time localize themselves within the map. The issue with single- robot SLAM is the relatively high rate of failure in a realistic application, as well as the time and energy cost. In this work, we propose a new approach to decentralized multi-robot SLAM which uses a robot swarm to map the environment. This system is capable of mapping an environment without human assistance and without the need for any additional infrastructure. We assume that 1) no robot possesses sufficient memory to store the entire map of the environment, 2) the communication range of the robots is limited, and 3)there is no infrastructure present in the environment to assist the robot in communicating with others. To cope with these limitations, the swarm system is designed to work as an independent entity. The swarm can deploy new robots towards the region that is yet to be explored, coordinate the communication between the robots by using itself as the communication network and replace any malfunctioning robots. The proposed method proves to be a reliable and robust exploration algorithm. It is shown to be a self-growing mapping network that is able to coordinate among numerous robots and replace any broken robots hence reducing the chance of system failure.

Large-scale scenarios such as search-and-rescue operations, agriculture, warehouse, surveillance, and construction consist of multiple tasks to be performed at the same time. These tasks have non-trivial spatial distributions. Robot swarms are envisioned to be efficient, robust, and flexible for such applications. We model this system such that each robot can service a single task at a time; each task requires a specific number of robots, which we refer to as 'quota'; task allocation is instantaneous; and tasks do not have inter- dependencies. This work focuses on distributing robots to spatially distributed tasks of known quotas in an efficient manner. Centralized solutions which guarantee optimality in terms of distance travelled by the swarm exist. Although potentially scalable, they require non-trivial coordination; could be computationally expensive; and may have poor response time when the number of robots, tasks and task quotas increase. For a swarm to efficiently complete tasks with a short response time, a decentralized approach provides better parallelism and scalability than a centralized one. In this work, we study the performance of a weight-based approach which is enhanced to include spatial aspects. In our approach, the robots share a common table that reports the task locations and quotas. Each robot, according to its relative position with respect to task locations, modifies weights for each task and randomly chooses a task to serve. Weights increase for tasks that are closer and have high quota as opposed to tasks which are far away and have low quota. Tasks with higher weights have a higher probability of being selected. This results in each robot having its own set of weights for all tasks. We introduce a distance- bias parameter, which determines how sensitive the system is to relative robot-task locations over task quotas. We focus on evaluating the distance covered by the swarm, number of inter- task switches, and time required to completely allocate all tasks and study the performance of our approach in several sets of simulated experiments.

Recently, there has been an increasing interest in robotics related to multi-robot applications. Such systems can be involved in several tasks such as collaborative search and rescue, aerial transportation, surveillance, and monitoring, to name a few. There are two possible architectures for the autonomous control of multi-robot systems. In the centralized architecture, a master controller communicates with all the robots to collect information. It uses this information to make decisions for the entire system and then sends commands to each robot. In contrast, in the distributed architecture, each robot makes its own decision independent from a central authority. While distributed architecture is a more portable solution, it comes at the expense of extensive information exchange (communication). The extensive communication between robots can result in decision delays because of which distributed architecture is often impractical for systems with strict real-time constraints, e.g. when decisions have to be taken in the order of milliseconds. In this thesis, we propose a distributed framework that strikes a balance between limited communicated information and reasonable system-wide performance while running in real-time. We implement the proposed approach in a game setting of two competing teams of drones, defenders and attackers. Defending drones execute a proposed linear program algorithm (using only onboard computing modules) to obstruct attackers from infiltrating a defense zone while having minimal local message passing. Another main contribution is that we developed a realistic simulation environment as well as lab and outdoor hardware setups of customized drones for testing the system in realistic scenarios. Our software is completely open-source and fully integrated with the well-known Robot Operating System (ROS) in hopes to make our work easily reproducible and for rapid future improvements.

Dans les environnements difficiles résultant de catastrophes naturelles ou d'accidents industriels, des robots mobiles peuvent être utilisés pour réduire les interventions humaines. Ces robots doivent pouvoir parcourir de longues distances, suivre des trajectoires précises, transporter des matériels et instruments, tout en étant robustes aux perturbations et aux défaillances éventuelles de leurs composants (capteurs, actionneurs). Dans cette thèse, nous considérons des systèmes composés de robots mobiles à deux roues motrices (2WD), reliés physiquement entre eux. Nous proposons des lois de commande permettant au système multi-robot de suivre une trajectoire de référence malgré la présence de défauts d'actionneurs. Différentes commandes tolérantes aux fautes (FTC : Fault Tolerant Control) sont proposées. Certaines sont des commandes dîtes passives, qui sont conçues pour être robustes à des défauts actionneurs sélectionnés, d’autres sont dîtes actives puisqu’elles intègrent un algorithme de diagnostic (observateur adaptatif non linéaire) qui détecte, localise et estime les défauts.Des résultats de simulation sont présentés tout au long de la thèse pour vérifier la validité et montrer les performances des algorithmes de commande tolérante proposés. / In harsh environments resulting from natural disasters or industrial accidents, reducing human interventions by increasing robotic operations is desirable. The main challenges to be considered are not only that the robots should be able to go over long distances and operate for relatively long periods, but also make the global system tolerant to actuators’ failures. In this thesis, to overcome these challenges, systems composed of multi-linked two-wheel drive (2WD) mobile robots are considered. The objective of these multi-robot systems is to asymptotically track a reference trajectory, despite the presence of actuator faults. In this thesis, we design original Fault Tolerant Control (FTC) schemes. Some of them are passive methods, i.e. robust control laws to given failures, and other ones are active FTC which include a Fault Diagnosis (FD) algorithm (nonlinear adaptive observer) that detects, localizes and estimates the faults, and finally adapt the control actions to the faulty situations. Simulation results are presented all along the thesis to verify the validity of the proposed control algorithms and to show the performance of the FTC schemes.

Despite the advancement of research and development on multi-robot teams, a key challenge still remains as to how to develop effective mechanisms that enable the robots to autonomously generate, adapt, and enhance team behaviours while improving their individual performance simultaneously. After a literature review of various multi-agent learning approaches, the two most promising learning paradigms, i.e., cooperative learning and advice sharing are adopted for future development. Although individually these methodologies may not provide a solution, their proper integration will provide a platform that allows for the incorporation of multi-agent learning with social behaviours. These methods are examined in relation to the performance characteristics of single-robot learning to ascertain if they retain viable learning characteristics despite the integration of individual learning into team behaviour. Further, various modifications to the Q-Learning algorithm were tested, and the best performing modification was implemented into the proposed multi robot learning approach.

Despite the advancement of research and development on multi-robot teams, a key challenge still remains as to how to develop effective mechanisms that enable the robots to autonomously generate, adapt, and enhance team behaviours while improving their individual performance simultaneously. After a literature review of various multi-agent learning approaches, the two most promising learning paradigms, i.e., cooperative learning and advice sharing are adopted for future development. Although individually these methodologies may not provide a solution, their proper integration will provide a platform that allows for the incorporation of multi-agent learning with social behaviours. These methods are examined in relation to the performance characteristics of single-robot learning to ascertain if they retain viable learning characteristics despite the integration of individual learning into team behaviour. Further, various modifications to the Q-Learning algorithm were tested, and the best performing modification was implemented into the proposed multi robot learning approach.

Robots have become increasingly adept at performing a wide variety of tasks in the world. However, many of these tasks can benefit tremendously from having more than a single robot simultaneously working on the problem. Multiple robots can aid in a search and rescue mission each scouting a subsection of the entire area in order to cover it quicker than a single robot can. Alternatively, robots with different abilities can collaborate in order to achieve goals that individually would be more difficult, if not impossible, to achieve. In these cases, multi-robot collaboration can provide benefits in terms of shortening search times, providing a larger mix of sensing, computing, and manipulation capabilities, or providing redundancy to the system for communications or mission accomplishment. One principle drawback of multi-robot systems is how to efficiently and effectively generate plans that use each of the team members to their fullest extent, particularly with a heterogeneous mix of capabilities. Towards this goal, I have developed a series of planning algorithms that incorporate this collaboration into the planning process. Starting with systems that use collaboration in an exploration task I show teams of homogeneous ground robots planning to efficiently explore an initially unknown space. These robots share map information and in a centralized fashion determine the best goal location for each taking into account the information gained by other robots as they move. This work is followed up with a similar exploration scheme but this time expanded to a heterogeneous air-ground robot team operating in a full 3-dimensional environment. The extra dimension adds the requirement for the robots to reason about what portions of the environment they can sense during the planning process. With an air-ground team, there are portions of the environment that can only be sensed by one of the two robots and that information informs the algorithm during the planning process. Finally, I extend the air-ground robot team to moving beyond merely collaboratively constructing the map to actually using the other robots to provide pose information for the sensor and computationally limited team members. By explicitly reasoning about when and where the robots must collaborate during the planning process, this approach can generate trajectories that are not feasible to execute if planning occurred on an individual robot basis. An additional contribution of this thesis is the development of the State Lattice Planning with Controller-based Motion Primitives (SLC) framework. While SLC was developed to support the collaborative localization of multiple robots, it can also be used by a single robot to provide a more robust means of planning. For example, using the SLC algorithm to plan using a combination of vision-based and metric-based motion primitives allows a robot to traverse a GPS-denied region.

In this thesis fundamental pieces of multi robot systems have been discussed and researched, in order to develop and build a system with easily obtainable electronics and to answer how much communication is needed as well as which design choices are important to make it robust. Our work will hopefully contribute to others in the community who are working with Raspberry Pi and Windows 10 IoT Core by being open source. As a result a proof of concept system of two simple robots has been implemented. In this paper we have used trigonometry and dead-reckoning for localization, when coordinating our robots a leader/follower model has been applied. The communication has been developed using the AllJoyn framework to develop an interface that has IoT capabilities. The results show that our system has working communication and simulated localization, however the limitations in the hardware results in an error in localization which we present in this paper. To answer our research questions the amount of communication needed is dependent on the problem and how many robots you need to apply in order to solve it and the most important design choice for current multi robot systems is a controlled environment.

Game theory can work as a coordination mechanism in multi-agent robotic systems by representing each robot as a player in a game. In ideal scenarios, game theory algorithms guarantee convergence to optimal configurations and have been widely studied for many applications. However, most of the studies focus on theoretical analysis and lack the details of complete demonstrations. In this regard, we implemented a real-time multi-robot system in order to investigate how game-theoretic methods perform in non-idealized settings. An aerial coverage problem was modeled as a potential game, where each aerial vehicle is an independent decision-making player. These players take actions under limited communication, and each is equipped with onboard vision capabilities. Three game-theoretic methods have been modified and implemented to solve this problem. All computations are performed using onboard devices, independent of any ground entity. The performance of the system is analyzed and compared with different tests and configurations

The frequent occurrences of natural and technological disasters have incurred grave loss of life and damage to property. For mitigating the miserable aftermaths, multi-robot teleoperation systems have been developed and deployed to cooperate with human rescuers in post-earthquake scenarios, and to sample, monitor and clean pollutants in marine environments. With a bidirectional communication channel, human users can deliver commands/requests to guide the motions of the remote robots, and can receive visual/audio feedback to supervise the status of the remote environment, throughout multi-robot teleoperation. Furthermore, the remote robots can send force feedback to human operators to improve their situational awareness and task performance. This way, a closed-loop multi-robot teleoperation system becomes bilateral in which coordinated robots physically interact and exchange energy with human users, and hence needs to be rendered passive for safe human-robot interaction. Beyond guaranteeing closed-loop passivity, the control of a bilateral multi-robot teleoperation system faces two challenging problems: preserving the communication connectivity of the remote robots; and dispatching the teleoperation authority to multiple human users. Because wireless transmission of radio/acoustic signals between the remote robots is constrained by their distances, bilateral multi-robot teleoperation control must coordinate the motions of the remote robots appropriately so as to maintain their communication network connected. Further, multiple human users can send possibly conflicting teleoperation commands to the remote robots, a distributed authority dispatch algorithm is thus needed for the remote robot network to recognize and follow the most urgent user commands at runtime. This thesis develops an energy shaping strategy to preserve the connectivity of the remote robots, and to dispatch control authority over the remote robots to human users, during bilateral multi-robot teleoperation. Chapter 1 introduces the application background of multi-robot teleoperation as well as the state-of-the-art development in related research areas. In Chapter 2, a dynamic interconnection and damping strategy is proposed to reduce and constrain the position error between the local and remote robots to any prescribed bound during bilateral teleoperation. Chapter 3 derives a gradient plus damping control from a bounded potential function and then unifies it into an indirect coupling framework to preserve all communication links of an autonomous multi-robot system with time-varying delays and bounded actuation. On these bases, Chapter 4 develops a dynamic feedforward-feedback passivation strategy to preserve all communication links and thus the connectivity of the tree network of the remote robots while rendering the bilateral multi-robot teleoperation close loop passive. Specifically, by blending the sliding variable in Chapter 2 with the bounded potential function in Chapter 3, the dynamic passivation strategy decomposes the dynamics of the remote robots into a power-preserving interconnection of two subsystems, and regulates the energy behaviour of each subsystem to preserve the tree communication connectivity of the remote robots. To handle time-varying communication delays, the strategy further transforms the communication channels between the local and remote robots into a dynamic controller for passivating bilateral teleoperation. Superior to existing controls, the strategy using a bounded potential function can circumvent numerical instability, reduce noise sensitivity and facilitate future extensions to accommodate robot actuator saturation. On the other side, Chapter 5 designs a distributed and exponentially convergent winners-take-all authority dispatch algorithm that activates the teleoperation of only human users with the most urgent requests in real time. After formulating the problem as a constrained quadratic program, we employ an exact penalty function method to construct a distributed primal-dual dynamical system that can solve the problem at an exponential rate. Because the equilibrium of the system changes with user requests, we then interconnect the dynamical system with physical robot dynamics in a power-preserving way, and passivate closed-loop multi-robot teleoperation using multiple storage functions from a switched system perspective. Finally, Chapter 6 provides some conclusive remarks and two problems regarding connectivity preservation and authority dispatch for future study. / Graduate