Planning and control for teams of robots in complex environments

We address the challenge of controlling a team of robots through a complex environment, a capability relevant to applications such as environmental monitoring or surveillance. The difficulty of the problem lies in the design of individual control laws for each of the agents that account for inter-agent interactions while preserving the convergence properties of the control law and drive the system to a desired state. We present an abstract statistical representation that reduces the complexity of controller design and is invariant to the number of robots in two and three dimensions. We develop control laws that require limited global state information, resulting in decentralized control laws at the agent level with collision avoidance and show convergence and stability properties. We propose estimation of the representation through the development of a distributed consensus algorithm and the inclusion of a supervisory aerial robot acting as a global observer. We define energy metrics based on the abstract representation that reduce the complexity of the planning problem to designing trajectories for the abstract representation of the system and discuss strategies for splitting and merging the robot formation. We review an experimental infrastructure developed to validate these multi-robot planning and control strategies and provide experimental verification on teams of nonholonomic robots. We conclude by proposing future research opportunities.