Integrated Design And Autonomous Capture Of Operational Aircraft Platform

In recent years,multi-rotor aircraft has achieved explosive development,such as the applica-tion of consumable drones in personal aerial photography and that of industrial drones in agricul-ture,forestry,logistics,meteorology,etc.Unlike the ordinary aerial vehicles,the latter requires a different type of aircraft to perform its tasks.This aircraft carries an operating mechanism that can exert an active influence on its environment.A number of research institutions have gradually carried out researches on the operational aircraft,including different types of flying robots with claws,arms and even multiple arms.Starting from the platform selection and mechanical design,we tried to sum up the basic needs of the ideal platform by iterative design in order to build an operational aircraft platform with a two-degree-of-freedom series manipulator.In the iterative design process,a set of competition platform with parallel manipulators was built for the completion of the seventh-generation mission in the IARC competition,and the top touch task of the ground robot was finally achieved by segmented trajectory planning.Through the dynamic modeling of the quadrotor platform,the attitude and position control from the bottom to the top is completed,along with a simple motion planning controller based on fixed-point control.At the same time,by combining the simple kinematics modeling,the forward and inverse kinematics solution of the manipulator as well as engineering practice,a complete me-chanical arm control plan including hardware interface,communication,application and controller layers,is thus obtained.The target capture of the aircraft has to be based on the relative position of the target and the aircraft.We conducted two sets of test experiments.First,target detection conducted by the OptiTrack motion capturing system.It is mainly operated by the motion capture device,the ground station software and the broadcast for real-time position detection.The second one is based on ArUco label detection algorithm whose target detection is completed by threshold filtering,contour filtering,bit extraction and ID recognition.The target tracking is controlled by additional feedback-corrected kernel correlation filtering algorithm while the relative pose solution of the aircraft and the target relies on PnP target solution.The second scheme becomes a set of stable target detection algorithms that can replace OptiTrack.Finally,in the experimental process,at first we built a simulation environment in the Unity simulation software for the previous structural design and later algorithm verification.Then we designed the target capture experiment in the real environment that includes the static target and the dynamic target capturing based on OptiTack and ArUco environments.The experiment verifies the flexibility and stability of the control algorithm and the target detection algorithm.Meanwhile,we also discovered some insufficiency of the traditional control methods and tried to solve the complex control problem of fast moving objects via deep reinforcement learning algorithm of proximal policy optimization.We use ML-Agents in Unity to build the training envi-ronment,complete the training,analyze the results,summarize the space for further modification and keep it for future improvement.