Fast Trajectory Planning for Multiple Site Surveillance through Moving Obstacles and Wind

In many civil and military applications, Unmanned Aerial Vehicles (UAVs) have to visit a partially-ordered set of strategic sites, often in presence of obstacles and wind. In this paper, we present a trajectory planning algorithm specially designed for this kind of mission, and possibly runnable during flight. The main idea of our approach is to proceed in three steps: paths building, sites ordering and velocity tuning. This separation allows to obtain a fast but incomplete method. Introduction Unmanned Aerial Vehicles (UAVs) are more and more used both in civil and military missions. Civil missions mainly divide into rescue and prevention tasks. In rescue operations, UAVs are sent to locate survivors in hostile environnements, for instance after a natural hazard. In prevention ones, UAVs keep a watch on a limited area. This area can be a piece of forest (to prevent fires) or sea (to prevent oil spills). Military missions consist in gathering data in enemy terrain (about troops, radars, warehouses, etc.). In addition to preserve pilot life, UAVs are accurate and often stealth. Thus, they represent a strategic benefit. These two types of mission can be seen as the partiallyordered surveillance of several sites, possibly in presence of obstacles and wind. Most of the existing planners can solve a part of this problem very efficiently. Only few planners can handle all the constraints, but they are quite slow. Consequently, we propose a planning method able to handle all the constraints quickly. This method is based on a decomposition of the initial problem into three sub-problems: the generation of the subpaths between sites, the ordering of the sites and the tuning of the UAV’s velocity. This decomposition allows to obtain good performances but leads to an incomplete method. State of the art In this section, we will first present the existing planners. Then, we will explain a potential field method, called wavefront expansion, from which the first step of our algorithm is inspired. The existing planners Most of the existing planners are designed to link two points in presence of possibly moving obstacles. This is a common problem in robotics: the robot has to reach a goal in a partially known environment. As the obstacles can be discovered during the exploration, the trajectory planning method has to be very fast. Generally, the space is discretized into a grid in order to apply discrete methods, as grid potential fields (Brock & Khatib 1999) (Kitamura et al. 1995) or variants of A* (Fraichard 1999). These efficient (re)planners deal neither with wind nor with multiple points. The planners which integrate (or could integrate) these constraints mainly use continuous optimization methods. Among them, we can find genetic algorithms (Torroella 2004) and mixed integer linear programming (Chaudhry, Misovec, & Andrea 2004) (Richards & How 2002). In both techniques, the computation time is too long to replan the trajectory during the flight. Indeed, the first technique requires a lot of generations, and the second one has to deal with a huge number of variables. The potential field Method The potential field method was introduced by Khatib for robotics applications (Khatib 1980). Indeed, it allows a robot to build a collision-free path from any point to a goal point among static obstacles. This is done in three steps. First, an attractive potential field is associated to the goal and repulsive ones to the obstacles. Then, the global potential field (correponding to the sum of all potentials mentionned above) is calculated (an example is given in fig. 1). Finally, the robot moves systematically towards the lowest values of the global potential.