Real-time Trajectory Planning For Quadrotors With Limited-range Sensing

Unmanned Aerial Vehicles(UAVs)have been widely used in tasks such as rescue and aerial reconnaissance because of strong maneuverability and low prices.Most missions are carried out in cluttered and unknown environment,so map is unavailable in advance.Therefore,UAVs have to perceive the environment with onboard sensors,and conduct real-time trajectory planning.This puts forward higher requirements for the autonomous navigation of the UAVs.Different from the well-developed trajectory planning in a known environment,the following issues need to be considered for real-time trajectory planning based on sensors with a limited range:(i)fast sensor data processing to construct a three-dimensional(3D)map in real time;(ii)fast obstacle avoidance during the flight;(iii)the ability to jump out of the “dead end”.To this end,we build a real-time trajectory planning system for quadrotors,which realizes incremental map construction and online trajectory generation.Receding horizon planning is used to design a re-planning scheme in order to solve the problem of real-time trajectory planning without prior environmental information and with only limited-range sensors.Firstly,starting from the mathematical model,the differential flatness of the quadrotor is derived,which provides a foundation for dimensionality reduction of trajectory planning.In the dynamic map generation part,the 3D occupancy grid map is researched.The map generation module is built through several steps including point cloud processing,raycast,and grid status update.The lidar is used to complete real-time construction test of the 3D grid map.Secondly,we focus on the 3D path search algorithm based on RRT* and the trajectory optimization method for constructing a quadratic programming problem.Path points are obtained by the search algorithm,then trajectory segmentations are generated according to the path points,finally they are combined under the re-optimization framework.The trajectory is expressed in polynomial form,so the trajectory generation is converted into an optimal solution of the quadratic programming problem.The simulation realizes the construction and verification of the trajectory planning module,and a comparative analysis of algorithms is carried out.Finally,a re-planning scheme is designed by the idea of receding horizon planning.Two sub-target selection methods are proposed,a linear heuristic one and a search heuristic one.Besides,a re-planning method combining period trigger and event trigger is adopted to improve the performance of the system.The system architecture and the way of integrating different modules are designed,and the verification system is built using ROS to verify and test algorithms in different environments.The results show that the system has good performance in obstacle avoidance,and the ability of solving local minimum problem,trajectory length and success rate of the algorithm are compared and analyzed.