3D SLAM With A Biped Climbing Robot In Truss Environment

Three-dimensional truss structure is a building structure made of columnar pole supported by each other.Compared with other types of robots,biped climbing robots（BiCRs）have unique advantages and more application potential in such environments.Point-to-point autonomous navigation is essential for efficient autonomous operation of BiCRs,in which the technique called simultaneous localization and mapping（SLAM）is the core par.However,existing SLAM schemes do not meet the needs of autonomous navigation for biped climbing robots.Therefore,this thesis proposes a SLAM method as well as system for BiCRs in truss environment.The main elements of this thesis are as follows.Based on the factor graph optimization theoretical framework,the mentioned SLAM method is represented as a factor graph containing the following types of factors: gripper factor,LiDAR odometer factor,forward kinematic factor,and pole point factor.In this thesis,the detailed derivation of the Jacobi matrix of the above-mentioned factors for the state variables provides a theoretical basis for improving the efficiency of solving the optimization problem.A pole landmark mapping method that consists of three parts: pole object perception,pole landmark data association and pole landmark parameterization.For the pole object perception,a multi-sensor fusion perception scheme is proposed,which can extract the points belonging to the pole from the LiDAR point cloud;for the parametrization of pole landmark,a nonlinear optimization problem is constructed to estimate pole landmark parameters;for the data association of the pole landmark,the axes of all pole landmark are discretized and the K-D tree data structure is used to save the discretized points,the data association is completed with the nearest neighbor search algorithm.The BiCR-SLAM system consists of four threads: LiDAR odometry thread,encoder thread,pole landmark mapping thread,and optimization thread.The encoder thread updates the graspping state of the robot and using positive kinematics to estimate the position of the end gripper in real time;the LiDAR odometry optimizes the pose of the end gripper and selects keyframes according to the pose difference;the pole landmark mapping thread implements the pole landmark mapping method proposed in this thesis;and the optimization thread optimizes the gripper poses and pole landmark parameters.The proposed SLAM scheme can meet the needs of autonomous navigation of the biped climbing robot,which establishs the technical foundation for the autonomous operation of the biped climbing robot in unknown environment,as well as effectively advances the process of replacing the human in the high-risk operation of the biped climbing robot in the truss enviroment.