Communications aware decentralized model predictive control for path planning within UUV swarms
Nicholas DiLeo, Alexandra Abad, Kingsley Pregene · 2017 IEEE Conference on Control Technology and Applications (CCTA) · 2017
We describe a decentralized model predictive control (DMPC) algorithm to control teams of unmanned underwater vehicles (UUVs) that simultaneously optimizes vehicle control inputs in a manner that explicitly accounts for the limitations of operating underwater, which include low bandwidth communications. Rather than treating the challenges of operating a swarm underwater, such as swarm communications and collision avoidance, as constraints to be satisfied, we formulate these factors as sub-objectives and include them directly in the optimization problem. This ensures that vehicles find a solution to the optimization problem and allows vehicles to dynamically prioritize subobjectives in situ. Swarm formations are therefore able to autonomously converge closer together in the presence of higher environmental noise. We also use a network graph to facilitate communications throughout the swarm. In contrast to existing methods that prescribe a maximum distance between all neighboring vehicles, our algorithm allows each vehicle to determine when it must limit its movement to keep communications throughout the swarm and when it can move more freely because it is less critical to the swarm's network graph. We also introduce the notion of a “critical node” within the swarm, which is a swarm member whose movement is restricted to keep in communication with its neighbors, maintaining a fully connected network graph. We demonstrate this scheme in several example scenarios where a team of UUVs works collaboratively to accomplish tasks while simultaneously addressing competing sub-objectives.