Research on Atutonomous Truss Assembly by Space Robot Using Object-Oriented World Model

对自主机器人视觉传感系统和空间机器人自主桁架装配进行了研究。研究结果总结如下：1.研究成果。研制了一种用于自主机器人的立体视觉系统。提出了一种基于层次形状图的有效分割方法，用于重建世界的三维结构。在目标识别中，利用CAD建立的数据库，提出了一种具有误差抑制的可靠模型匹配方法。为了实时观察目标的运动，采用单视觉系统的高速图像处理算法来计算目标的位置和姿态。通过并行编程将两个过程相结合，实现了实时的运动跟踪。提出了一种新的分割方法来区分物体的曲线轮廓。在极线约束下，对左右图像之间的对应曲线进行立体匹配。然后，重建物体的三维几何特征。在模型匹配中，比较重建对象和数据模型的大小和形状信息。研究了基于空间机器人的桁架结构自主装配。实验气台系统在地面上实现了二维无摩擦状态。立体摄像系统按照规定的桁架装配顺序，识别目标和障碍物的位置和姿态。然后，通过势法生成机械手的路径，以获得避免与障碍物碰撞的手的期望位置和姿态。将桁架的基础结构视为障碍物，机械手可以将每个桁架单元携带到基础结构上而不会发生碰撞。利用机械手上的力矩传感器进行力控制，实现了机器人的顺利装配。通过硬件实验验证了桁架结构自主装配的可行性。

A visual sensing system for autonomous robot and the autonomous truss assembly by a space robot are studied. The results are summarized as follows.1. A stereo vision system for an autonomous robot is developed. An effective segmentation method using a hierarchical shape graph is introduced to reconstruct the 3D structure of the world. In the identification of the targets, using database built by CAD, a reliable model matching method with error rejection is proposed. To observe motions of the objects in real-time, high-speed image processing algorithm by a single vision system is used to calculate the position and posture of the targets. Combining both processes by parallel programming, the real-time motion tracking is realized.2. A new segmentation method is proposed to distinguish a curved outline of the object. The stereo matching is performed for the corresponding curves between left and right images under an Epipolar constraint. Then, the three-dimensional geometrical feature of the object is reconstructed. In the model matching, information of the size and the shape is compared between the reconstructed object and the data model.3. An autonomous assembly of truss structures using an space robot is studied. An experimental air-table system realizes a two-dimensional frictionless condition on the ground. According to the prescribed order of the truss assembling, the stereo camera system recognizes the positions and attitudes of the target and the obstacles. Then, the manipulator path is generated by a potential method to attain the desired position and attitude of the hand avoiding collision with obstacles. Considering the base structure of the truss as an obstacle, the manipulator can carry each truss element to the base structure without collision. The smooth assembling is realized by the force control using torque sensors on the manipulator. The feasibility of the autonomous assembly of the truss structure is demonstrated by the hardware experiment.