Multi-Robot Nonlinear Model Predictive Control for Persistent Monitoring

We present an approach to address a multirobot persistent monitoring problem, where a team of agents must repeatedly survey specific points of interest (POIs) within a 2-D area. Our approach models the interest value of each POI with a heat-like dynamics. Each agent then solves online a nonlinear model predictive control (NMPC) problem to determine feasible trajectories that minimize the cumulative heat across all POIs. The trajectories are parameterized with Bézier curves, whose control points are used as optimization variables; this parametrization enables agents to efficiently communicate their optimized motions. An additional quadratic optimization layer adds safety guarantees while a central unit updates the global POIs’ map. The method has been validated in simulation and real experiments, demonstrating that the algorithm can run online and on computationally limited hardware platforms. In addition, an extensive simulation campaign compares our NMPC against a state-of-the-art baseline across 90 randomly generated scenarios with different numbers of POIs. Our NMPC outperforms the baseline along the considered metrics, attaining lower robot velocities.