Coordination of multirobot teams and groups in constrained environments: Models, abstractions, and control policies

Robots can augment and even replace humans in dangerous environments, such as search and rescue and reconnaissance missions, yet robots used in these situations are largely tele–operated. In most cases, the robots\u27 performance depends on the operator\u27s ability to control and coordinate the robots, resulting in increased response time and poor situational awareness, and hindering multirobot cooperation. Many factors impede extended autonomy in these situations, including the unique nature of individual tasks, the number of robots needed, the complexity of coordinating heterogeneous robot teams, and the need to operate safely. These factors can be partly addressed by having many inexpensive robots and by control policies that provide guarantees on convergence and safety. In this thesis, we address the problem of synthesizing control policies for navigating teams of robots in constrained environments while providing guarantees on convergence and safety. The approach is as follows. We first model the configuration space of the group (a space in which the robots cannot violate the constraints) as a set of polytopes. For a group with a common goal configuration, we reduce complexity by constructing a configuration space for an abstracted group state. We then construct a discrete representation of the configuration space, on which we search for a path to the goal. Based on this path, we synthesize feedback controllers, decentralized affine controllers for kinematic systems and nonlinear feedback controllers for dynamical systems, on the polytopes, sequentially composing controllers to drive the system to the goal. We demonstrate the use of this method in urban environments and on groups of dynamical systems such as quadrotors. We reduce the complexity of multirobot coordination by using an informed graph search to simultaneously build the configuration space and find a path in its discrete representation to the goal. Furthermore, by using an abstraction on groups of robots we dissociate complexity from the number of robots in the group. Although the controllers are designed for navigation in known environments, they are indeed more versatile, as we demonstrate in a concluding simulation of six robots in a partially unknown environment with evolving communication links, object manipulation, and stigmergic interactions