Micron is an actively stabilized instrument for microsurgery. Cam Riviere has been working on this in his lab since he came to CMU in the 90's, but I was a key contributor to getting the concept to actually work to any notable degree while handheld, which requires precise high bandwidth control in at least three degrees of freedom.

The 3DOF Micron system is described in this paper. This was the first version of Micron that demonstrated quantifiable accuracy improvement in a multi-subject trial, and I was involved from the beginning of this version. As well as the reimplemented ASAP tracker, there was a new manipulator based on a unique bending mode actuator.

I came up with a flexure-based manipulator using a folded pair of these piezos to increase the range of motion. This particular arrangement also happens to provide a better combination of range of motion and stiffness than the stacking arrangements that the manufacturer suggests. Greg Podnar did the mechanical design, and I came up with the somewhat involved assembly procedure, using various kinds of adhesives, conductive epoxy, etc., and several machined alignment fixtures. I used a flex PCB to make the connections to the piezos and to the LEDs used for position tracking. There are no active electronic components in the 3DOF handpiece.

I wrote the control software (running at 2K sample/sec) using Labview real-time. I also designed the multi-subject experiment used to evaluate human-in-the-loop performance and analyzed the data using n-way ANOVA (in MATLAB.)

This version of Micron changed to a 6DOF configuration and smaller overall size. This is the first Micron where we conducted in-vivo tests (in rabbit). See this paper for some details. The ASAP position tracker component is basically the same, but the manipulator is all new, and considerable control development was necessary as well.

Greg Podnar and I came up with the basic manipulator concept, but Sungwook Yang took over the design and assembly, including some very nice kinematic and static analysis, described in the paper. Kinematically, the manipulator is a hexapod, and a quite small one at that. The overall diameter is less than 25mm.

It uses 1.8mm Squiggle motors (from Newscale technologies). These piezoelectric linear motors allow a more compact handpiece, while giving an arbitrarily large range of motion. We designed for a 10x larger range of motion than in the 3DOF system (4mm as opposed to the 400 microns we were getting) because the 3DOF range of motion was barely adequate for tremor cancellation, frequently saturating even if you had a fairly steady hand. A larger range of motion allows improved tremor suppression by larger scaling factors, and also permits autonomous scanning for Optical coherence tomography and Laser photocoagulation.

The 6DOF motion is also important because it allowed us to implement a remote center of motion, which is necessary in the eye-surgery scanning modes, since the tool shaft must pass through front part of the eye (see Vitrectomy). 6DOF motion is also beneficial in basic cancellation because it allows us to stabilize the entire tool pose, and not rely on being able to compensate lateral translation with tip/tilt motions.

The Squiggle motors are another unique actuator. In comparison to conventional linear ultrasonic motors, they get a considerable mechanical advantage by using the ultrasonic resonance to rotate a screw. This allows a much larger output force in a given volume, but creates awkward problems because the output screw must be free to rotate, and the motor nut is not designed to serve as a linear bearing. Usually the output shaft simply presses on a bearing surface, which in turn slides on a linear guide. We chose to use the motors with no linear bearing to save space and reduce mechanical complexity. Theoretically in a hexapod there is no lateral force on the links because they have ball joints. Our flexures have some stiffness, creating a side load on the motors, but the design it seems to work acceptably, thought the motors make a lot more squeaking noise than in the recommended application mode.

There were considerable hardware challenges in design and assembly of the handpiece because of the need to integrate the motor driver chips into the handpiece. I used a stack of 20mm diameter boards. Three identical driver boards stack, each rotating 60 degrees so that the flex connectors face in an appropriate direction. Below that is a differential line driver/receiver board, and a board with a voltage regulator and connectors for the LED flex, and then a board that transitions from the cable to the stacking connectors. The connectors are 0.5mm pitch, the drivers a 4x4mm QFN, and the small passive components are 0402. It isn't practical to hand-solder these components, so I used a solder stencil, then hotplate reflow on the topside components and a hot-air rework tool to reflow the bottom side.

Upgrading the Labview code to operate in 6DOF operation was fairly easy, but getting satisfactory control of the rather weak and unreliable motors in the presence of mechanical imperfections proved challenging. Though Sungwook did a modal analysis of the manipulator, which suggested adequately high first resonance (~200 Hz), it wasn't practical to model the nonlinear backlash and wobble of the screw in the motor nut. We found that the motor performance varies significantly between assemblies, and the velocity resulting from a given motor command is significantly non-linear. I addressed this by a calibration procedure that does a polynomial fit. But there were more complex dynamic control issues that were not entirely resolved.

Statically the backlash in the link assembly is taken up by a preload spring, but when the dynamics are high this force can be overcome, revealing the backlash. The problem is exacerbated by the relatively narrow lean angle of the links, which we chose to minimize the overall diameter. This reduces the stiffness of the lateral translation and axial rotation modes. I try to keep the manipulator out of this chattering mode by limiting the commanded accelerations, and using a multi-model Kalman filter for the link length to identify when the link is misbehaving, reducing the gain to get it out of chattering.

The approximately 40 micron p-p chattering at the tip looks ugly, but actually doesn't have a huge effect on the achievable static positioning accuracy that we measure in our human-in-the-loop experiments. The larger lower frequency components of tremor dominate.