视觉是赋予机器人以智能的重要途径，而视觉伺服控制则是机器人利用视觉信息灵活完成复杂任务的一种重要方法。近年来，尽管骨科手术机器人凭借其在精度、透视辐射剂量和微创程度相比传统手术的优势得到了越来越多的临床应用，但历经数十年的发展后现有的大多数产品仍然局限于基于光学标志跟踪的技术框架。光学标志跟踪技术提供了机器人和手术对象的绝对空间坐标，为机器人技术的应用提供方便的同时也带来了坐标变换链复杂，操作流程繁琐以及二次创伤等问题。受空间交会对接任务中航天器之间采用相对位置制导方式的启发，本文提出将视觉伺服控制方法应用于机器人骨科手术，通过图像信息直接获取手术器械和手术对象的相对位置关系，以此代替光学标志反馈机器人的运动。为此，本文在对现有的视觉伺服控制方法进行发展的基础上，研究了视觉伺服控制在两种典型骨科手术操作中的实现问题。首先针对螺钉置入手术抽象而来的空间射线对准的视觉伺服控制问题进行了研究。对于三维图像引导的情形，提出了一种基于运动学关系的“相机-机器人”自标定方法，并分别给出了时间连续和离散情形下三维视觉伺服的控制稳定性条件。对于二维图像引导的情形，提出了射线位姿解算“共线保持性”的概念，并基于此提出了一种混合视觉伺服算法，在抵消标定误差影响的同时避免了二维视觉伺服的奇异性问题。对于以上两种情形，分别给出了三维结构光和二维X光透视引导的机器人螺钉置入的具体实现方法。其次，研究了骨折复位手术抽象而来的包含柔性环节的二维视觉伺服问题，并针对其特点提出了一种“状态依赖”的加权最小二乘雅可比矩阵估计方法，以及一种基于几何参数的雅可比矩阵初始化方法，用于补偿柔性环节的影响。在此基础上给出了一种二维透视引导下机器人骨折复位的实现方法。对于上述三种具体的手术应用场景，分别设计并搭建了对应的手术机器人原型机，通过模型或动物尸体实验对本文所提出的视觉伺服方法进行了验证。结果表明，在不依赖额外光学标志的情况下，借助视觉伺服方法仍可以使手术精度满足临床要求，这充分体现了通过视觉伺服控制方法取代光学标志跟踪技术的可行性。
Vision information provides robots with intelligence, and more specifically, visual servo control is an important way for robots to accomplish complicated missions. The clinical application of orthopedic surgical robots is becoming more and more widespread in recent years due to their superiorities of better accuracy, less radiation dose and less invasiveness compared with traditional non-robotic surgery. However, even after decades of development, most orthopedic surgical robots still rely on the marker-based optical navigation system. Such optical navigation system provides absolute spatial coordinates of the robot and the surgical subject, which provides great convenience for robot control, while also raises problems like error accumulated along complex coordinate transformation chains, tedious operation procedures, and extra iatrogenic injuries.Inspired by the relative position-based guidance strategy used in the rendezvous and docking of spacecrafts, in this paper we studied the robotic orthopedic surgery problem under the visual servoing framework, through which extra optical markers are no more necessary. In such a framework, we directly obtain the relative position between the surgical tool and the surgical subject through intro-operative images, with which the robot is controlled in a feedback manner. Specifically, we developed the visual servo control methods and studied their applications in two typical orthopedic surgeries.Firstly, the screw placement operation was abstracted out as a visual servo control problem of aligning two spatial half-lines. When 3D images are used, we proposed a self-registration method based on kinematic constraints to obtain the “camera-robot” transformation, and derived the stability criteria of the 3D visual servo controller for both continuous-time model and discrete-time model. When 2D images are used, we proposed a concept called “collineation retainability”, based on which a hybrid visual servoing method is developed. This hybrid method can not only compensate the calibration error but also avoids the singularity of the 2D visual servoing method. As for implementations, a 3D structured light guided and a 2D X-ray guided robotic screw placement system were realized, respectively.Secondly, the 2D visual servo problem with flexible components, which was abstracted out from the fracture reduction surgery, was studied. A “state-dependent” weighted least square Jacobian estimator, and a geometry-based Jacobian initializer were developed to compensate the deformation in the system. Then an intraoperative 2D X-ray guided robotic fracture reduction system is proposed.For each of the three applications mentioned above, we built a prototype system and tested the visual servoing methods on models or cadavers. The results show that with the proposed visual servoing methods the surgical accuracy can meet the clinical requirements, even without using extra optical markers, which demonstrated the feasibility of replacing the marker-based optical navigation technique with the visual servo control method.
