SPARTAN

Aerial systems with suspended payloads enable time-critical delivery in unstructured and hard-to-access environments.Onboard visual navigation remains challenging due to coupled robot-payload dynamics, limited field of view, and the need for reorientation around large obstacles. Imitation learning can enable fast onboard navigation under partial observations by distilling payload-aware, globally informed planning behaviors. However, its use in aerial transportation has been limited by the lack of scalable tools for generating diverse, dynamically feasible, and long-horizon expert demonstrations. To address this gap, we present SPARTAN, a framework that couples a scalable, perception-aware global expert planner with a domain-randomized tracking controller to generate large-scale, diverse datasets for vision-based policy learning. Our method jointly accounts for obstacle avoidance, complex payload dynamics, and field-of-view alignment while supporting configurable dataset generation across varied experimental settings. Using demonstrations collected across 110,000 environments, we present, to the best of our knowledge, the first large-scale data-generation pipeline for vision-based aerial transportation with suspended payloads, and show that policies trained on its data reliably support navigation in challenging cluttered environments.