On Communication and Flocking in Multi-Robot Systems

Abstract: Coordination of multi-robot systems to improve communication and achieve flocking is the topic of this thesis. Methods are proposed for mobile autonomous robots to follow trajectories in a way that improves communications with a base station. Further, a decentralized algorithm is presented that yields flocking with obstacle avoidance.The communication-aware trajectory tracking is adapted to radio communication in indoor environments. Our experimental data show that the effect of multipath fading, well-known in the radio communication literature, causes significant variations in the signal strength between a mobile robot and a base station. A contribution of this thesis is to formulate a tradeoff between tracking a reference trajectory and maintaining communication, first for a stationary reference position and then for a general trajectory. For the general case, the robot and an onboard communication buffer are modelled as a hybrid system, switching between standing still to communicate at positions with good signal strength and driving to catch up with the reference. This problem is solved using relaxed dynamic programming. For the case of a stationary reference position, experimental validation shows that loss of communication is avoided and that the method yields a gain in signal strength.The algorithm for flocking is based on Voronoi partitions. They can be approximated using only local information and allow the agents to avoid collisions. Our contribution is to add obstacle avoidance and movement towards a goal by using a navigation function - a scalar potential field with exactly one local minimum at the goal. To bound the inter-agent distances and thus avoid flock dispersion, any agent on the boundary of the flock uses a mirroring mechanism to create virtual neighbors that drive it inwards. We can prove collision safety and bounded group dispersion, and simulations show reliable goal convergence even in the presence of non-convex obstacles. A version of the algorithm with lower computational complexity is also presented. It can be used for formation control and it is proven to be locally asymptotically stable for a particular case. A hierarchical control structure is proposed for implementing the flocking on non-holonomic vehicles. It has been tested on a realistic car-like robot model in a flight dynamics simulator and the results confirm that the results on safety, group dispersion and goal convergence apply also in this case.