MPPI Control for Physical Human-Drone Interaction

In recent years, the collaboration between human operators and UAV systems has introduced significant challenges due to the inherent uncertainty in human behavior. This thesis presents a novel approach to account for such uncertainty in a human-drone interface system by leveraging a Model Predictive Path Integral (MPPI) Control technique. The controller has been designed to work for a collaborative payload transportation between human and UAV. In particular the task is to transport a load (represented by a bar) between two points. The human model can describe a sequence of forces applied by the human to the object during the task. The probabilistic model can describe a trajectory through mean and variance, allowing the MPPI controller to generate more coherent trajectories for the human dictated by the probabilistic model. To accurately create the human's model in this framework, a dedicated dataset was collected and used to estimate the governing non parametric Stochastic Differential Equation, enabling dynamic prediction and adaptation to human actions. The resulting controller enhances the drone's knowledge of the human movements making it cooperative by effectively accounting for the uncertainty inherent in human input.