Connectivity-Aware Model Predictive Control for Swarms of Dynamically Constrained Vehicles Engaged in Multi-Target Persistent Observation

Maintaining a connected communication graph while simultaneously observing spatially distributed targets is difficult for vehicles that cannot stop, reverse, or rotate in place. We introduce and compare five controllers for this mission: centralized and hybrid heuristics, a static convex optimizer, and centralized and distributed Model Predictive Control (MPC). The formal predictive controllers optimize curvature and acceleration actions directly, propagate a nonlinear constrained-vehicle model, and constrain predicted communication connectivity. We then examine how the coordination principles transfer to six forward-only ground vehicles. The robot implementation uses a convex kinematic receding-horizon surrogate followed by deterministic, iterative waypoint-separation repair. For distributed MPC, one fixed-identity agent runs on each robot computer, solves a local optimization problem onboard using previous-cycle peer predictions, and returns a waypoint proposal to a centralized safety supervisor. The five controllers are evaluated in 50 physical trials, ten per controller. Over a common 60 s window, centralized MPC achieved the lowest integrated mission cost, which was 6.0% smaller than that of the static convex controller and 52.2% smaller than that of the hybrid heuristic. Distributed MPC achieved the largest mean algebraic connectivity but incurred a higher mission cost. The centralized heuristic attained the smallest priority-target distance by explicitly favoring that target, while paying a larger total cost through weaker coverage of secondary targets. The experiments reveal a performance-connectivity trade-off and show that waypoint-level separation is not a continuous collision-avoidance guarantee.