【作者】 张国亮；

【导师】 刘宏；

【作者基本信息】 哈尔滨工业大学 ， 机械电子工程， 2010， 博士

【摘要】 利用空间机器人实现航天器的在轨服务代表了空间服务技术的发展趋势,但是,受到当前传感信息理解、机器人运动控制以及人工智能等技术的制约,完全自主工作的智能空间机器人在短时间内还难以实现。因此,目前在轨的空间机器人主要采用地面遥操作控制。但是,传统的遥操作控制方法受系统时延影响较大,且对于环境的适应能力也较差,这直接影响了空间机器人遥操作系统的性能。因此,研究视觉技术与遥操作结合的控制方法对于克服遥操作通讯时延、提高空间机器人的自主作业能力、扩展空间机器人的应用范围具有十分重要的意义。为此,在国防科工委“卫星在轨自维护及遥操作关键技术的研究”项目的支持下,论文对空间机器人遥操作中基于视觉的局部自主进行了研究。围绕空间机器人遥操作系统中的时延问题,论文的研究内容主要从两个方面展开:一方面,利用虚拟现实技术和实时视觉反馈信息辅助操作者进行遥操作,提高了传统主从式遥操作的操作性能;另一方面,为彻底消除时延对遥操作系统的影响,对基于视觉的空间机器人局部自主操作当中的目标识别以及视觉伺服进行了深入细致的研究。论文建立了卫星自维护遥操作系统的模型,实现了基于虚拟现实的三维图形预测仿真,有效地克服了时延对遥操作系统的影响。针对视频传输过程中波动带宽与视频流码率的不匹配问题,提出分层码率控制的带宽自适应策略,实现了遥机器人远端场景的视频反馈,提高了遥操作系统的视觉临场感。为实现空间机器人的局部自主操作,分别从结构化环境和非结构化环境两个方向考虑,重点研究了用于实现自主操作的目标识别以及视觉伺服方法。针对结构化环境,设计了结合矢量图压缩与形状拟合的目标识别方案,通过对目标角点特征的亚像素优化提取,实现了模型化目标准确、稳定的识别。根据模型化目标的视觉信息,探讨了目标位姿的非线性优化计算方法,并根据目标位姿计算的结果,提出了利用在线估计目标的深度和旋转姿态来选择PBVS和IBVS控制器的复合视觉伺服方法,保证了空间机器人自主操作的伺服精度和伺服范围。通过引入冗余自由度机器人的冗余空间自运动特性,有效地克服了典型IBVS方法的摄像机退化问题。对于非结构化环境,为提高目标识别对环境变化的适应性,提出一种分布式多视觉特征融合的目标识别策略。该方法通过对每个特征子集的模糊化加权融合处理,实现了目标在复杂的动态环境中的鲁棒提取和跟踪。针对非结构环境下空间机器人遥操作系统的作业问题,提出利用操作场景的视觉动态检测建立任务层观测器,并结合操作者的抓取经验知识和机器人的局部自主进行遥操作控制的方法,提高了遥操作系统的操作性能以及处理意外事件的能力。最后,利用卫星自维护遥操作地面实验平台,对文中提出的方法进行了实验验证。对于结构化环境,应用复合视觉伺服控制方法,完成了大范围插入探针的实验,有效地扩展了卫星自维护空间机器人的自主操作空间。对于非结构化环境,利用操作场景的视觉动态检测结果在任务层进行规划,实现了擦拭卫星镜头的实验。更多还原

【Abstract】 Utilizing space robot to realize on-orbit servicing of spacecraft represents space servicing technical trend. However, it is difficult to develop an autonomous intelligent space robot in a short time due to the limit of current sensor comprehension, control of robot motion and artificial intelligence. As a result, current on-orbit space robot is mainly controlled in teleoperation method. However, long time delay and unadaptive to environment is a general obstacle in traditional teleloperation, which directly affects performance of space telerobotics system. Therefore, it is significant for space robot to study integration of vision technology and teleoperation, which not only overcome teleoperation time delay, but also improve autonomy performance and enlarge application. This dissertation focus on local autonomy based on vision in space robot teleoperation under surporting of the project“study on the key technologies of satellite on-orbit self-servicing and its teleoperation”.This dissertation can be mainly classified into two parts for disposing of time delay in space robot teleoperation system. On one hand, from the view of improving traditonal teleoperation performance, a vision aid teleoperation is investigtaed by combining virtual reality and real vision data. On the other hand, object recognition and visual servoing in local autonomy of space robot is detailly investigated in order to completely cut off time delay in teleoperation system.Satellite self-serving teleoperation system is established in this dissertation. The three-dimensional graphic predictive display is realized by virtual reality to overcome the influence of time delay on teleoperation. In order to deal with mismatch of bandwidth and video stream bit rate, a bandwidth adaptive scheme based on scalable bit rate control is proposed to realize video feedback of remote robot part. Accordingly, visual telepresence experience of teleoperation system is highly improved.To accomplish local autonomy of space robot in structured environment and unstructured environment, this dissertation focuses on object recognition and visual servoing of local autonomy. For the structured environment, a model-based object recognition is investigated based on vector compression and shape approximation. Furthermore, sub-pixel of corner feature is extracted and optimized to realize precise and stable recognition of model-based object. Utilizing vision data of model-object, object pose is estimated based on non-linear optimization. According to on-line estimation of depth and rotation pose of object, a hybrid visual servoing method is proposed to guarantee precision and range in local automation of space robot. Based on characteristic of self motion in redundant freedom robot, a partitioned visual servoing is realized to handle camera retreat in image-based visual servoing.For the unstructured environment, in order to improve the adaptability of vision system for various environments, an object recognition strategy based on multi distributing visual feature integration is proposed, which implement robust recognition and tracking of feature in complex and dynamic environment. To realize space robot teleoperation under unstructured environment, a teleoperation strategy is presented based on combination task-level observer built by dynamic visual inspection, operator grasping experience and local autonomy of remote robot, which improve the systeme operation performance and capability of handling accident.At last, theory and method proposed in the dissertation are tested by experiment based on the ground teleoperation demonstration system of satellite self-serving. In the structured environment, a probe insert experiment under wide range is carried out with hybrid visual servoling, which enlarge automation operation space of satellite self-serving system effectively. In the unstructured environment, utilizing task-level regulation, the experiment of rubbing camera lens are realized based on dynamic visual inspection of operation scene.更多还原