Model Predictive Path Integral Control Framework for Partially Observable Navigation: A Quadrotor Case Study

Recently, Model Predictive Path Integral (MPPI) control algorithm has been\nextensively applied to autonomous navigation tasks, where the cost map is\nmostly assumed to be known and the 2D navigation tasks are only performed. In\nthis paper, we propose a generic MPPI control framework that can be used for 2D\nor 3D autonomous navigation tasks in either fully or partially observable\nenvironments, which are the most prevalent in robotics applications. This\nframework exploits directly the 3D-voxel grid acquired from an on-board sensing\nsystem for performing collision-free navigation. We test the framework, in\nrealistic RotorS-based simulation, on goal-oriented quadrotor navigation tasks\nin a cluttered environment, for both fully and partially observable scenarios.\nPreliminary results demonstrate that the proposed framework works perfectly,\nunder partial observability, in 2D and 3D cluttered environments.\n