| With the rapid development of micro unmanned aerial vehicle(UAV)and more extensive application scenarios,autonomous navigation of rotor UAV has gradually become a research hotspot at home and abroad.The autonomous navigation of rotor UAV is evaluated by the fusion of Global Positioning System(GPS),lidar,vision sensor and inertial sensor.The diversity of the scene makes the rotor UAV rely on GPS and visual navigation failure in the case of poor GPS signal and poor illumination.However,lidar can achieve high-precision pose estimation without relying on the GPS and illumination of the surrounding scene.Therefore,the navigation system based on lidar has a wider range of scenarios and is more suitable for autonomous navigation of micro UAV.Based on the premise of independent of GPS and illumination,this paper focuses on the positioning technology and local path planning algorithm of rotor UAV based on multi-source sensor fusion such as lidar.The research contents of this paper are as follows:(1)According to the existing mature driverless technology,a rotor UAV autonomous navigation platform architecture is constructed.The three-layer architecture of physical layer,protocol layer and application software layer is designed.The UAV,sensors and other hardware equipment needed to realize the navigation system are selected.The rotor UAV autonomous cruise system is realized by combining hardware equipment and software architecture,It includes control system of ground station,cruise system,positioning system and flight control system of rotor UAV(2)Based on the rotor UAV platform,a multi-source sensor localization method based on Unscented Kalman Filter(UKF)algorithm is proposed.Firstly,the mainstream positioning technology of autopilot is introduced.According to the research direction and task of the project,after comparing the advantages and disadvantages of the existing lidar point cloud registration algorithm,the 3D Normal Distributions Transform(NDT)algorithm with higher registration efficiency and lower accuracy is selected.Aiming at the problem of registration accuracy,considering the application of sensor aided positioning,the positioning accuracy is improved by fusing the data of inertial measurement unit with the pose of point cloud after registration.Considering that the motion model of rotor UAV is a nonlinear model,UKF combined with Constant Turn Rate and Velocity(CTRV)motion model is used to fuse lidar data and inertial measurement unit data.Aiming at the abnormal data of lidar and inertial measurement unit caused by the instability of rotor UAV in the air,an online positioning evaluation method is proposed to evaluate and recover the positioning failure.Through the design of the whole positioning module,the navigation accuracy and stability of the rotor UAV in flight are ensured.(3)In the case of the existing map,the unknown obstacles will consider all kinds of situations,resulting in complex logic execution.According to the actual situation of rotor UAV,this paper proposes a reinforcement learning local path planning algorithm based on strategy search,designs the motion and state space of rotor UAV,reduces the dimension of 3D lidar data,and designs the reward function and experience playback function in line with the cruise scenario.Experiments show that this method can make the rotor UAV avoid local obstacles and reach the designated target point at the same time.(4)The fusion localization experiment and autonomous flight experiment are designed and verified in the stadium.The experimental results show that the fusion localization algorithm can effectively realize the positioning of rotor UAV and ensure the successful completion of cruise mission;The simulation environment of reinforcement learning obstacle avoidance scene is constructed,and the effectiveness is verified in the simulation environment.The experiment shows that the local path planning algorithm can complete the obstacle avoidance task and reach the target point. |
PDF Full Text Request