Model predictive UAV-tool interaction control enhanced by external forces

Abstract

One of the major challenges of automated systems is attributed to the interaction task. This process involves external forces which may be dangerous, for example in an unmanned aerial vehicle (UAV) that interacts with unknown environment. There are numerous potential applications in UAVs that require physical interaction with its environment. However, this scenario brings evident challenges to be addressed, i.e., (i) the dedicated parts for the physical interaction (e.g., robotic arm) might change the moment of inertias and the center of gravity of the UAVs; (ii) contact phase might cause chattering effects; (iii) versatile external forces during interaction can degrade the performance; (iv) the needs of the UAVs to respect the bounds on the controller actions as well as the upper limits of the additional sensors equipped with a tool. In order to handle the aforementioned challenges in a systematic way, an optimization–based approach is proposed for use on the control and the estimation design. The translational states and unmeasured forces are estimated by nonlinear moving horizon estimation (NMHE) after each new measurement becomes available. The estimated external forces are then fed into the nonlinear model predictive control (NMPC) which provides the total force and three angular positions. For the total forces, a novel control allocation is designed to maintain the interaction with the ceiling at the desired level. The angular values, three outputs of NMPC, are given to proportional–derivative–integral (PID) controllers to maintain attitude stability. Using the external force information given by the NMHE, the presented interaction controller is able to interact with the environment experimentally in milliseconds.