Inferring the interaction topology in agent-based network systems from observable data is a fundamental challenge in a wide variety of disciplines, including Multi-Robot Systems, Neurobiology, Social Sciences. We study the statistical learning framework for multi-robot systems to infer the underlying interaction relations, given data consisting of sampled trajectories of interacting robots. We demonstrate the effectiveness of our approach by providing theoretical guarantees and testing it on multi-robot rendezvous, formation and swarming systems.