Cooperative UAV Formation Flying With Obstacle/Collision Avoidance

Navigation problems of unmanned air vehicles (UAVs) flying in a formation in a free and an obstacle-laden environment are investigated in this brief. When static obstacles popup during the flight, the UAVs are required to steer around them and also avoid collisions between each other. In order to achieve these goals, a new dual-mode control strategy is proposed: a "safe mode" is defined as an operation in an obstacle-free environment and a "danger mode" is activated when there is a chance of collision or when there are obstacles in the path. Safe mode achieves global optimization because the dynamics of all the UAVs participating in the formation are taken into account in the controller formulation. In the danger mode, a novel algorithm using a modified Grossberg neural network (GNN) is proposed for obstacle/collision avoidance. This decentralized algorithm in 2-D uses the geometry of the flight space to generate optimal/suboptimal trajectories. Extension of the proposed scheme for obstacle avoidance in a 3-D environment is shown. In order to handle practical vehicle constraints, a model predictive control-based tracking controller is used to track the references generated. Numerical results are provided to motivate this approach and to demonstrate its potential.

I. Introduction

There has been a spurt of interest in recent years in the area of unmanned aerial vehicle (UAV) formation control and obstacle avoidance. Many interesting applications of formation flying have been studied. Examples include forest fire monitoring [1], radar deception [2], and surface-to-air missile (SAM) jamming [3]. Main considerations in formation flying include how to come together and maintain a formation and how to achieve collision/obstacle avoidance. Obstacle avoidance in a 2-D environment has been studied earlier [4], [5]. In [4], the concept of rapidly exploring random trees (RRT) originally investigated in [6] was used to find dynamically feasible obstacle-free paths. After obtaining an obstacle-free path, a reactive path planner was used to avoid pop-up obstacles. This generates trajectories that are not taxing for the vehicle, i.e., the trajectories involve less turns. As the main idea is to track a path generated a priori and avoid pop-up obstacles as they appear, this scheme does not guarantee a shortest distance path always. Also, the RRT scheme requires larger computation times than the method proposed in this brief. In [5], mixed integer linear programming (MILP) was used to design dynamically feasible trajectories for obstacle avoidance. However, the MILP scheme requires larger computation capabilities when compared to the scheme proposed in this study. Furthermore, the 2-D approach presented in this brief gives the shortest distance paths.