Learning-Based Methods for Aerial Manipulation: A Focused Review

Learning techniques are increasingly empowering aerial robots with manipulation capabilities and have, as a result, attracted growing attention over the past decade. This work provides a focused review of recent advancements in learning-based methods for aerial manipulation (AM). Four categories of learning-based methods, i.e., imitation learning, reinforcement learning, deep learning, and learning-based control, are identified and analyzed in detail. The AM tasks enabled by these methods are also summarized and analyzed, resulting in six categories: Grasping, pick-and-place, load transportation, contact-rich operations, contact impact, and human-UAV interaction. In addition, we provide further discussions and perspectives on the current achievements and future directions of learning-based AM.