Interleaved predictive control and planning for an unmanned aerial manipulator with on-the-fly rapid replanning for object retrieval in unknown environments

Unmanned Aerial Manipulators (UAM) are gaining attention within the unmanned aerial systems research community. They can be used for aerial manipulation tasks such as object retrieval from confined and hard-to-reach spaces. The coupled dynamics between the arm and the base and also the limited flight time would require the development and implementation of optimal motion planning and robust flight control strategies. In this work, we propose a novel integrated planning and control strategy for object retrieval. A new kinodynamic version of the RRT*, called Lazy-Steering-RRT*, is developed for planning UAM's motion from its start to a pre-grasp state, while keeping the motion of the arm to a minimum. This planning can be carried out on the fly by using Machine-Learning-based techniques to construct the edges in the search tree in a time-efficient way. This facilitates re-planning, as the environment is gradually sensed by limited range sensors onboard. Once the UAM reaches the pre-grasp state at the end of the motion cycle, an RRT approach is then utilized, where motions of the base and the arm are coordinated for reaching and grasping the object. An MPC/feedback-linearization control approach is also utilized for end-effector trajectory tracking based on a novel cross-dynamic partitioning approach between the arm and the base. The overall motion planning and control algorithms have been implemented in simulation using Multibody Physics Engines for realistic results and a number of representative simulation runs are presented. Our results show that the approach is effective in successfully executing the object retrieval task even in presence of confined spaces.