Abstract:
Rapid advances in sensing, computing and communication technologies have led to considerably increased research activities in multi-robot systems over the last decade. Topics include multi-robot motion planning, cooperative manipulation, aerial applications involving cooperative exploration of the unknown environment, automated highway systems, software architectures for multi-robot systems, and formation control. Multi-robot systems have been proven to offer additional advantages in terms of flexibility in operating a group of robots and failure tolerance due to redundancy in available mobile robots. However, the benefits of using multi-robot teams do not come without cost. Coordinating teams of autonomous robots is much more challenging than maneuvering a single robot.
This dissertation addresses formation control problems, which are among the most active research topics in multi-robot systems. Over the last two decades, there have been a large number of publications on this field, and it is still growing. Recently, this research has been extended to some related research areas, e.g., consensus problems and distributed control systems, imposing new challenges on formation control problems. In general, formation control subproblems addressed in the literature can be classified as formation shape generation, formation reconfiguration/selection, formation tracking, and role assignment in formation. The main purpose of this dissertation is to address two important and correlated subproblems in formation control: formation tracking and role assignment in formation. The goal of the former is that a team of mobile robots is required to maintain a geometric formation while tracking a reference or a set of references. The latter arises when a mobile robot in the team must decide what role to take on in a desired formation configuration.
In particular, we study coordinated path following control of omnidirectional mobile robots and unicycle mobile robots. This problem can be seen as a subtask of formation tracking. Path following is one of the three basic motion control tasks in mobile robot research. The others are trajectory tracking and point stabilization. Even though less attention is drawn to this problem in the literature, it offers some advantages over trajectory tracking in some cases. The objective of path following control is to be on the path rather than at a certain point at a particular time. To solve this problem, we employ a model predictive control (MPC) technique to generate a sequence of optimal velocities of a so-called virtual vehicle which is followed by a real robot. This approach can eliminate stringent initial condition constraints because the velocity of a virtual vehicle is controlled explicitly. Using this technique, we can gain some benefits over other available control schemes, e.g., the ability to incorporate generic models, linear and nonlinear, and constraints in the optimal control problem and the ability to use future values of references when they are available, allowing to improve system performance. However, the main drawback is significant computational burden associated with solving a set of nonlinear differential equations and a nonlinear dynamic optimization problem online.
Then, we extend path following control to coordinated path following control. A group of mobile robots not only follow a reference path but also maintain a geometric formation shape. The main challenge is to design a decentralized control law using only local information to achieve a formation tracking objective. In this study, we propose two solutions. In the first solution, the MPC framework for path following control is extended to the coordinated path following control problem. In spite of great theoretical properties of such MPC controllers, the stability and feasibility of decentralized schemes are rather conservative. The second solution is computationally simple so that it may be suitable for low-computational systems when the advantages of MPC schemes including constraint handling are not a dominating factor. Its controller design is based on a Lyapunov technique and a second-order consensus protocol with a reference velocity. It is worth noting that the path variable has been used as a coupling variable synchronizing each member in formation in both solutions.
In the second formation control subproblem, we study role assignment in formation. This problem becomes more challenging when robots in the team do not have complete information and they do not know the number of robots participating in the formation tasks. With the assumption that the formation graph is connected and bidirectional, we propose an online and distributed role assignment. This approach is proven by extensive simulation and experimental results.