Research On Pose Estimation And Motion Control Of Aerial Bionic Neck-eye System

Micro unmanned multirotors have been widely investigated due to its simple structure and model,small size and strong mobility.In most applications,two basic capabilities are necessary,i.e.,the perception ability and planning and control ability.Traditionally,the constraints of obstacle avoidance and motion models are taken into account in the trajectory planning of drones,and perception is tightly related to the type,quantity and distribution of image sensors and actual scenes.However,sensors and the body of the vehicles are often strap-down in most aerial vehicles.This leads to the serious coupling between the received information and the planned trajectory.For example,the camera information is sensitive to observation angle,moving speed and external environments,therefore increasing the risk of perception system failure.The strap-down mechanism prevents efficient combination of the two capabilities to satisfy the requirements of many practical tasks.To address the above practical problems,the thesis focuses on improving the stability and robustness of the perceptual localization system.Inspired by the birds' vision system,a novel bionic aerial mechanical neck-eye system is designed.A dual-fisheye inertial module is equipped at the end of the 6-Do F neck,treated as the perception module of the bionic system.To estimate the vehicle pose,a pose estimator is developed based on the pose propagation from the perception module.In this way,the perception module can be controlled by the mechanical neck,and it can move relatively independently to its body.By separating the sensor from the vehicle body,the flexibility of vehicle motion is improved and the ability to active perception is enhanced.The hardware of the bionic aerial mechanical neck-eye system is designed by constructing the multirotor body,the neck mechanism and the design of the dual-fisheye inertial module.The forward and inverse kinematics of the mechanical system are modeled.And then the transformation between the perception module and the vehicle body is derived.Based on the calibrated fisheye model,the visual strategy of the dual-fisheye module was proposed.Several virtual pinhole cameras are generated by using the fisheye camera image.Further,a filter-based tightly coupled visual odometer using local sliding window optimization is proposed.The odometry enhances the linearity and consistency of the system by using inverse depth parameterization and robot-centric representation.Thephotometric errors between image patches from consequent image frames are integrated as the innovation of the extended Kalman filter.State quality of the filter is monitored for guaranteeing the robustness of the pose estimator.The accuracy of the estimator is further improved by optimizing the states within a predefined sliding window.Based on the state estimator and the dual-fisheye perception strategy,a scene richness model is further constructed and an active perception planning algorithm for the system is proposed.A minimum snap optimization method to improve the smoothness of the planning result is performed,utilizing the closed-form solution for calculation efficiency.With the above solution,the coupling between the motion of the mechanical neck and vehicle body is reduced.To verify the feasibility of the bionic aerial neck-eye system,the calibration of the fisheye is first validated,and then the pose estimation algorithm is evaluated based on both the public dataset and the real platform.Simulation and physical platforms are also built to verify the effectiveness of the active perception algorithm in ensuring the robustness of the aerial system.The mechanical neck-eye system is a novel solution of localization,planning and control for unmanned aerial vehicles.The proposed system has potential applications in many fields with unmanned aerial vehicles.