A

D. VISION, CONTROL AND TOOLING SUBSYSTEMS

This thesis is concerned with the mechanical aspect of building a CABG surgery platform. However, we will briefly discuss the necessary requirements for vision, control and tool subsystems needed for the actual working device.

The task of the Stewart Platform manipulator is to carry the camera and the motorized surgical instruments maintaining a constant pose with respect to the heart. That can be achieved by controlling the manipulator closed loop, by means of vision and co ntrol subsystems. The vision subsystem provides the relative position of the heart relative to the actuated platform. The control subsystem takes that data and calculates the next position of the platform. This feedback from the vision subsystem should be generated in real-time and contain 6-DOF positioning information. The choice of the vision sensors depends on the required accuracy and system bandwidth.

Since the surgeon can operate using the input from monocular or stereoscopic images that are stabilized, the same images can be used for the feedback. The problem of maintaining the constant pose is then expressed in terms of vision data, and it is cal led image based servoing. This method operates in the following way. The first image of the tracked object is taken and a reference feature vector is defined from this image. The feature vectors are then calculated from all subsequent images of the moving object. The difference between the reference and the new feature vectors defines the error vector. From the error vector, the controller derives the position error and generates appropriate trajectory. To enhance the visual tracking accuracy and speed, s pecial markers are usually placed on the tracked object, which can be imaged robustly. The markers can be color coded [Wei97] to allow fast computation of position and orientation of the heart. The number of markers should be redundant in case that some o f them are occluded by the surgical instruments or bleeding from the surrounding tissue.

The bandwidth of the system has to be sufficient to follow the motion of the heart. That means that the manipulator has to be faster than the beating heart and that the sampling rate of the vision system needs to be high. The faster alternative to the off-the-shelf video cameras is the microsonometry system. They are small sonic transducers that can be placed on the surface of the heart and on the platform. For example, ultrasonic spatial localization device Sonic Digitizer GP8-3D (Science Accessories Corp., Stratford, Conn.) could determine the positions of ultrasound transmitters in space, sampling the distances between the transmitters at the rate of 24kHz [Reinhardt96].

The control subsystem has the task of generating the trajectory of the Stewart Platform manipulator. The heart motion is fast and the delays of the platform and the vision subsystem have to be eliminated. The real-time coordination of a robotic manipul ator with a moving target is described in [Allen93]. The tracking is performed there using optical flow and predictive Kalman filtering to null out video processing delays. The motion of the heart is approximately repetitive and periodic. That makes the r epetitive control laws well suited for this problem. The repetitive and learning controllers use the information from the previous periods to calculate the trajectory in the future. For example, a simple learning control algorithm could be used [Phan88]:< /P>

u_k+1(t) = u_k (t) + F · e_k(t)

The command signal u_k+1(t), in the (k+1)-th repetition is adjusted proportional to the error e_k(t) in the k-th repetition, F is a gain matrix. Using learning control, periodic motion of the heart can be tracked with zero time lag [Arimo93], which is not possible to achieve using causal filters. The start of each repetition can be accurately determined from the heart ECG signal or from the external source (pacemaker) if the heart is paced.

With respect to tooling, different types of telerobotic surgical instruments have been implemented. Our device requires custom made telerobotic end-effectors that will be small, rigid and lightweight. The instruments need to be small and rigid to minim ize oscillations due to the movement of the platform. Cable driven tools, as described in [Schener95] would not therefore be appropriate, and magnetically levitated master-slave surgical instruments [Salcu97] are too massive. An interesting design of clos ed link master-slave surgical manipulator is described in [Mitsu97]. This manipulator is driven directly by an electromotor. For the CABG surgery device, a similar design can be employed, but the motors should be positioned off the platform in order to de crease the instrument weight.