A Distributed Autonomous System for Multi-UAVs With Limited Visualization: Employing Dual-Horizon NMPC Controller

Autonomous cooperative flight of multiple unmanned aerial vehicles (UAVs) in complex and unknown environments, such as jungles, poses significant challenges. Existing approaches often rely on wireless communication or omnidirectional vision to achieve cooperation. This article presents a dual-horizon nonlinear model-predictive control (NMPC) scheme for vision-limited multi-UAV systems in cluttered scenarios. The proposed scheme serves as a local path planner for UAVs equipped with forward vision limited to a 160$^\circ$ field of view, enabling distributed cooperative flight, obstacle avoidance, and navigation. To expand the observation area, UAVs randomly oscillate left and right around their direction of motion. The NMPC employs an objective function incorporating distributed cooperative flight and obstacle avoidance considerations during navigation. This objective function consists of short-term reactive objectives to maintain formation and consistent distance from neighbors and long-term predictive objectives for planning collision-free trajectories. CasADi, an open-source tool for nonlinear optimization and algorithmic differentiation, is utilized to solve the NMPC problem. The trajectory calculation is performed using initial conditions and fed as additional input to the NMPC. The first control signal of the predictive control sequence is then transmitted to the autopilot for UAV control. The proposed NMPC approach provides real-time solutions with a 10-Hz operating frequency and a 2-s prediction horizon, making it suitable as a local path planner. The effectiveness of the proposed approach is evaluated in three challenging scenarios using Gazebo-ROS.