I presented initial work on a clinical-scale interleaved continuum-rigid manipulator at the IEEE conference on Intelligent Robots and Systems (IROS 2014) in Chicago this past September.
Interleaved Continuum-Rigid Manipulation
In this paper we continued to develop the concept of Interleaved Continuum-Rigid Manipulation -- first described at ICRA 2013 -- by building a 5 degree-of-freedom prototype manipulator.
We built this prototype to demonstrate that rigid joints can be realized on a clinically-relevant scale and that multiple rigid joints can be used to correct for the error in multiple flexible segments. This post and the IROS 2014 proceedings paper are primarily focused on this first, design part, while our ICRA 2015 paper evaluates a controller that can realize the error correction.
Focusing on the design, in contrast to our previous single degree-of-freedom testbed this prototype demonstrates that interleaved manipulation can be realized in a clinically-relevant form factor. Manual and robotic catheter systems used in hospitals to treat atrial fibrillation are approximately 4mm (5/32") in diameter, so this is the target. The following picture shows a detail of the distal sections with a #2 pencil for comparison; the prototype catheter section is 6.35mm (1/4") in diameter which grows to 13mm (1/2") at the rigid joints and proximally. This was designed in approximately a month, and with that little amount of effort the prototype is 3x larger than clinical devices. There are many sub-optimal elements of this design, but we're in the ballpark. As discussed later, the driving element of this design is the planetary gearhead selection (one is visible as a black cylinder book-ended by stainless steel bearings). Their diameter directly leads to the 13mm device diameter, so by a trivializing extension gearheads with diameters of 1.5 mm (the size of a grain of rice) would lead to an approximately 3.5mm clinical scale device. While this hand-wave argument has ignored many other aspects of the design, it should suffice to show that the development prototype is a reasonable trade between a clinical device and one that can be cheaply and quickly produced in-lab.
Degrees of Freedom
As seen in the composite figure, the manipulator has four actuated joints: a base roll, distal pitch, distal roll, and catheter articulation. These joints are actuated by motors located outside of the notional patient. Their motions are communicated to the distal joints by flexible transmissions which are able to negotiate a flexible vasculature model. The catheter actuation is decoupled from the rigid joint motion by routing the actuating tendon through the rigid joint axes. See the above video to better appreciate the joint motions.
Rigid Joint Design
There is a large number of possible joint arrangements, designs, and actuation technologies that could be applied to an interleaved manipulator. The defining element of this prototype is the use of remote actuation with large, local reductions. Whereas most gear boxes are mounted onto the driving motor, the 700:1 planetary gearheads used in this prototype are located in the distal joints. They are connected to the motors by thin, piano-wire driveshafts which are able to bend with the vasculature model. This flexibility leads to a difference in rotation between the end attached to the motor and the other end attached to the planetary gearhead, but these errors are minimized by that large ratio (1/700). The result is that the distal joints are not impaired by the unknown number of twists and bends in the vasculature model. (The same cannot be said for the two flexible joints - base roll and catheter - as they do not have any local reduction and therefore respond to every twist and bend of the vasculature model.)
The design of both rigid joints is driven by the gearhead selection; in the above figure they are the adjacent gray cylinders in the vertical cross-section. These are 6mm in diameter with molded-plastic gears and carriers. Their inputs are driven by the flexible driveshafts at several hundred revolutions per minute, while outputs drive a cable assembly which actuates the joints. The cable assembly was judged more compact and easier to fabricate than other designs involving gears or levers. Though difficult to appreciate without a 3D model or lab tour, the following gallery shows some different views of the joints.
Outside of the gearheads, shafts, bearings, and pulleys, the rigid joints were 3D printed via stereo lithography at 10μm resolution. We chose to 3D print early in the design process, allowing much of the complexity to be put into the 3D printed parts and simplifying those that needed to be produced traditionally. Even given this, fabrication of millimeter-sized pulleys and shafts was difficult and the assembly much more so.
Our interest in this paper was showing redundant motion, where the same task can be performed by the two flexible segments or the rigid joints. The two demonstrations were conducted without any feedback control; a 6DoF electromagnetic tracker located at the catheter tip recorded the manipulator pose, but the manipulator commands were based solely on the forward kinematics. For the rigid joints to be able to correct for the flexible errors, the flexible and rigid workspaces must overlap. In the simple case, the task is designed to be planar such that no motion is required from the base or distal rolls.
The error grows throughout this task, as might be expected with open loop control. Also, as detailed in the introduction and by Mike and Jinwoo there are a variety of nonlinear effects which are not included in the forward kinematics here.
The highlight video above includes the 3D results video.