Embodied AI for aerial manipulation: a survey

Existing surveys on aerial manipulation mainly focus on platform design, modeling, control, and applications, while the recent transition toward embodied intelligence remains insufficiently structured. This survey reviews embodied intelligence for aerial manipulation from the perspectives of manipulation tasks, aerial embodiments and intelligent methodologies. First, aerial manipulation tasks are classified according to their interaction characteristics, ranging from grasping to contact-rich manipulation. For aerial embodiments, we categorize aerial vehicles and end-effectors, followed by a review of representative works. From the intelligent methodological perspective, this survey provides an in-depth analysis of four paradigms: model-driven, human-input-driven, data-driven, and hybrid model-data-driven methods. For each paradigm, we examine its technical principles, representative implementations, applicable tasks, and current limitations. In addition, recent progress in swarm embodied aerial manipulation is discussed. Finally, some open challenges and future research directions are identified to provide insights into the development of more autonomous, intelligent, and robust aerial manipulation systems for field application.