Autonomous Path Planning For Micro Aerial Vehicles In Outdoor Unstructured E Nvironments

Autonomous navigation task of micro aerial vehicles(MAVs)in outdoor unstructured environments with complex obstacles and uncertain structure is a challenging work,which includes a series of complex sub-tasks such as the motion control of MAVs' simultaneous position and mapping and path planning.The paper conducts research on the autonomous path planning for MAVs,including specific problems such as path finding,trajectory optimization,autonomous hover and flight strategy selection.This paper proposes a low-power quadrotor navigation system that can be used in unknown outdoor environments.In order to reduce the computational burden of drone,a separate architecture is designed to transfer computing resources from the onboard computer to the ground station.The communication bandwidth between both platforms is guaranteed by the 5G wireless network.Considering the payload and the endurance of the quadrotors,a binocular stereo vision camera is selected as the front-end sensor of the quadrotor to sense the surrounding environment and a dense point cloud map is built based on obstacles in the surrounding environment.Then point cloud data is converted into octree maps for the quadrotor's trajectory planning.The trajectory of the quadrotor is planned in two stages.In the first stage,a path finding algorithm is improved for the quadrotor to find paths in real 3D environments.Firstly,the algorithm can obtain a series of collision-free safe spaces on the octree map.Secondly,these safe spaces are expanded in the same direction as the coordinate axes until they reach the scene boundary or obstacle boundary.Finally,a series of overlapping flight corridors are generated.In the second stage,multi-segment Bezier curves are used to represent the trajectory and the trajectory generation problem is modeled as a quadratic programming problem.A safe and smooth trajectory which meet the dynamic feasibility of quadrotors is generated by adding a series of constraints.Because the application scenario of the drone in this experiment is an outdoor unstructured environment,and the effective farthest sensing range of the visual sensor is 8m,the drone can not directly obtain the prior map containing all the obstacle information of the scene.Therefore,the exploration in outdoor environments such as woods are full of unknowns.For solving this problem,an optimistic flight strategy is adopted.First of all,the areas which are outside the currently perceived range of the drone are considered as obstacle-free.Then as the drone continues to explore the unknown environment,more obstacles are added to the perception map.Once the collision between the trajectory tracked by drone and an obstacle is detected,a new trajectory would be re-planned to bypass the obstacle,immediately.This re-planning mechanism can guide the drone following the trajectory until the target position is reached.In order to ensure the safety of the implementation of the entire solution,the DJI assistant 2 software is used to simulate M100 quadrotor for simulation testing,and the drone flight status data is sent to the RVIZ visualization platform in real time for observing the tracking effect of the aircraft.We verify the feasibility of the system in a complex woods scene.The final results show that the scheme we designed can complete functions such as simultaneously position and mapping,path planning,and tracking flight in unknown unstructured scenes.