Autonomous UAV Navigation in an Unknown Environment via Multi-Trajectory Model Predictive Control

Abstract

A novel model predictive control (MPC) formulation, named multi-trajectory MPC (mt-MPC), is presented and applied to the problem of autonomous navigation of an unmanned aerial vehicle (UAV) in an unknown environment. The UAV is equipped with a LiDAR sensor, providing only a partial description of the surroundings and resulting in time-varying constraints as the vehicle navigates among the obstacles. The control system layout is hierarchical: the low-level loops stabilize the vehicle’s trajectories and track the set-points commanded by the high-level, mt-MPC controller. The latter is required to plan the UAV trajectory trading off safety, i.e. to avoid collisions with the uncertain obstacles, and exploitation, i.e. to reach an assigned target location. To achieve this goal, mt-MPC considers different future state trajectories in the same Finite Horizon Optimal Control Problem (FHOCP), enabling a partial decoupling between constraint satisfaction (safety) and cost function minimization (exploitation). Recursive feasibility and, consequently, persistent obstacle avoidance guarantees are derived under the assumption of a time invariant environment. The performance of the approach is studied in simulation and compared with that of a standard MPC, showing good improvement.