Study On Visual Navigation And Control For An Intra-vehicular Assistant Robot

Intra-vehicular assistant robot is an intelligent vehicle that operate inside the space station to assist astronauts in performing in-orbit missions and carrying out scientific experiments.This paper mainly focuses on the navigation and control problem of the assistant robot.Visual navigation approach is proposed to realize the intra-vehicular navigation,and multi-propeller thrusters are applied to achieve six degrees of freedom control of the robot.The main work of the dissertation is as follows:Aiming at the special application background that the space station environment is relatively stable and the operation range of the robot is limited,the visual navigation method based on sparse point cloud map is proposed.Sparse point cloud map is compact in expression,efficient to compute,and can support drift-free position determination with centimeter-level accuracy and attitude determination with 1 degree accuracy,which can meets the navigation needs of the assistant robot.Considering different orbital operation stages of the assistant robot and the changes of the cabin environment,a targeted visual navigation strategy is adopted.In the initial stage and when the environment changes obviously,the construction of sparse point clouds map and visual navigation are carried out simultaneously.In later stage,and when the environment is stable,the task of map construction is stopped and the assistant robot make use of the constructed map for navigation to reduce calculation consumption and avoid over-saturation of the map.Aiming at the specific task of simultaneously localization and mapping in the initial stage,problems of feature extraction and matching,visual odometry,map construction and optimization,and pose determination and optimization based on constructed map are thoroughly studied.The computationally efficient ORB features with scale,rotation and illumination invariance is proposed for navigation,and the uniformity of the feature in the extraction process is improved.A method of using priori motion information to guide feature matching and to enhance the matching efficiency and accuracy is proposed.An approach of nonlinear pose optimization based on unconstrained Lie algebra is given.The relative position and attitude motion model of the intra-vehicular assistant robot and the space station was established.An integrated six-degree-of-freedom position and attitude control of the assistant robot based on a set of propeller thrusters and PD controller is realized,and the asymptotic stability of the close-loop control process is proved.By analyzing the dynamic characteristics of the relative motion of the robot in phase plane,a sliding mode variable structure control method with quadratic type switchover surface is proposed.The proposed sliding mode controller is able to overcome the adverse effects of airflow interference,modeling error and measurement noise to the linear feedback controller and the overshoot and steady state errors that occur in the PD control process is eliminated,and the robustness of the control is improved.Sparse point cloud map,dense point cloud map and octomap of the space station ground simulation cabin is constructed using RGB-D data from Kinect depth camera.The validity of the simultaneously localization and the mapping method in the initial stage of operation is verified.A visual navigation ground experiment system based on quadrotor is established.Based on the quadrotor and the constructed sparse point cloud map,the automatic return experiment that the quadrotor reaches a predetermined target point from random initial position and attitude is realized,and thus the navigation method based on constructed map in the later stage of the assistant robot is verified.