Issue 30 · Pick 08 Robotics ✓ read
Eversion-based robots can enable safe access,steering and endoscopic imaging within the spinal subarachnoid space
A 2 mm tube that navigates the fluid-filled space around your spinal cord not by being pushed in, but by growing from its tip like a vine—so the part already inside never slides against the fragile nerves it passes. In a phantom this cut mean tissue-interaction forces by 65%, and in an intact human cadaver it reached 150 mm up the spinal canal from a normal lumbar puncture, with a chip-on-tip camera streaming the whole way and no visible damage afterward.
Why pushing a catheter up the spine is the wrong tool
Picture threading a wet noodle up a long, soft, crowded tube. Every clinical catheter and guidewire works by proximal pushing: you shove from the base, and that force has to travel all the way down the shaft to move the tip. Along the way the entire length of the device drags against the wall.
In most of the body that's tolerable. In the spinal subarachnoid space it is not. This is one of the narrowest, most delicate corridors in the body: the lateral gap between spinal cord and dura averages only ~2.5 mm per side, the dura itself is ~0.3 mm thick, and the corridor is packed with nerve roots and thin connective trabeculae, all bathed in cerebrospinal fluid. Two things go wrong as you push deeper. Accumulated friction along the shaft means less and less of your push actually reaches the tip, so you lose control exactly when you need it. And that same friction is shear force applied directly to neural tissue.
This is why intrathecal drug delivery—injecting therapeutics straight into the CSF, bypassing the blood-brain barrier—is dramatically underused. The paper cites England, 2020: ~450 patients considered for intrathecal delivery for cancer pain against an estimated 8,000+ eligible. Deep, controlled navigation past the initial lumbar entry is just too risky and too hard.
Continuum robots and steerable catheters help with flexibility, but they still push from the base. The core mechanical problem is untouched.
The idea: move only the tip, keep the body still
Eversion (vine) robots change what "advancing" means. Instead of sliding a shaft forward, you have a thin tube that is folded inside-out, stored in a pressurized chamber. Raise the internal pressure and the tube turns itself inside out at the tip, laying down new material at the very front. The crucial consequence: the outer surface of everything already deployed stays stationary relative to the surrounding tissue. There is no sliding along the shaft, because the shaft doesn't move—it only extends at the frontier.
Vine robots aren't new—Hawkes' 2017 work started the field, and there have been demos in colons, airways, and blood vessels. Two things make this paper more than a new application. First, the scale and integration: a 2 mm PTFE sheath (20 µm wall) carrying a coaxial steerable, imaging-capable instrument—a 1 mm² NanEye chip-on-tip camera, an illumination fiber, and a NiTi tendon for steering, all in a 1.45 mm inner instrument. Second, and more interesting, the mechanics of making that work at this scale turned out to defy the textbook.
The surprise: at 2 mm, friction rewrites the kinematics
Here's the wrinkle. If you have an inner instrument (the camera/steering catheter) living inside the everting sheath, the two must stay tip-to-tip aligned, or you lose your view and your steering. The authors run a four-phase "duty cycle": (1) pressurize, (2) advance—the inner catheter moves forward and the sheath everts at half the catheter's speed, (3) depressurize, (4) retract the catheter back to re-align its tip with the sheath tip, then repeat.
Idealized kinematics gives a clean prediction. Since the sheath grows at half the catheter's advance, retracting the catheter by exactly half its advance (a retraction ratio r=0.5) should restore alignment every cycle. And dropping chamber pressure to atmospheric (~100 kPa) should release the coupling so the catheter can slide back freely.
Both predictions fail. At r=0.5 the instrument tip consistently lagged the sheath tip, the offset accumulated cycle after cycle, the sheath scrunched at the distal end, and it eventually ruptured. The reason is that sheath–instrument coupling is dominated by frictional hysteresis, not ideal kinematics—residual friction persists even at atmospheric pressure, so "commanded symmetric motion" does not produce symmetric displacement.
The fix is the genuinely instructive part. They stop treating r as a geometric constant and start treating it as a friction-compensation knob: r=0.45 (retract slightly less) plus a sub-atmospheric low-pressure phase (P_l=90 kPa, below the 100 kPa ambient). The partial vacuum presses the folded sheath layers together and pins the deployed sheath in place, so the inner catheter can retract independently without dragging the sheath backward. With P_h=140 kPa, P_l=90 kPa, r=0.45, alignment held over the full 150 mm. This is the kind of "why it actually works" mechanism that transfers to anyone building miniature soft-growing devices: pressure isn't just the growth actuator, it's a real-time clutch controlling sheath–instrument coupling.
The numbers: force reduction and where it comes from
The phantom study is the quantitative backbone. Three matched strategies on an anatomically scaled, transparent cervical-spine phantom mounted on a six-axis force sensor: continuous conventional insertion, duty-cycle insertion (same timing/velocity, without eversion—so it isolates the growth mechanism), and full Vine4Spine.
The telling comparison is duty-cycle vs. eversion. Just chopping motion into duty cycles lowers the mean force (67 vs 100 mN) but does nothing for peaks (258 vs 243 mN)—because each little advance still relies on pushing. Only eversion cuts both: 65.2% off the mean and 48.0% off the peak versus conventional, and it produces the smoothest force profile. Axial mean force drops 79.8%. The reductions show up in transverse components too, so it's not just the insertion direction.
Why does growth do this? A nonlinear finite-element model (LS-DYNA) reproduces the force-reduction ratios well (simulated 48.6% mean / 55.8% peak vs measured 65.2% / 48.0%) and reveals what the load cell can't see. A pushed catheter concentrates stress at the advancing tip and nearby nerve-root attachments—peak von Mises stress 1.95 MPa. Eversion spreads contact along the stationary deployed sheath (contact area up to ~12.5 mm² vs a small localized patch), dropping peak stress to 0.39 MPa—about 5×—and cutting mean interfacial shear stress ~92% and peak shear ~86%. The mechanism is load redistribution, not merely a lower total force. (Caveat the authors flag: absolute peak forces are overpredicted, e.g. 419 mN vs measured 243 mN for conventional, from uncertain friction coefficients and the fact that the load cell measures a transmitted reaction force, not interface force. Treat the FEA as mechanistic, not predictive.)
The cadaver: the part that hadn't been done before
The headline demonstration is a fresh-frozen intact human cadaver (male, 92, no spinal pathology), accessed the ordinary way—fluoroscopy-guided lumbar puncture with a Tuohy needle and a clinical introducer, no laminectomy for access. This matters: the prior "thecaloscope" work that navigated the intrathecal space required surgical exposure and was fully manual.
They preloaded ~50 mm of sheath in the introducer, kept saline flushing at the entry to lubricate and prevent air ingress, and mixed contrast agent into the pressurization fluid so the everting sheath was independently visible under fluoroscopy. The robot then grew its full 150 mm—~80 mm of it free intrathecal navigation beyond the introducer—across multiple vertebral levels, teleoperated by joystick with simultaneous fluoroscopic and endoscopic views. Steering authority survived full deployment: tendon actuation still produced ~12 mm of controlled tip displacement at maximum extension. The pressurized chamber stayed sealed even at 180 kPa working pressure (burst-tested to 190 kPa), with no contrast leakage.
Afterward, neurosurgeons did a laminectomy and durotomy to look directly: no macroscopic disruption of dura, no tearing, deformation, or contusion of cord or nerve roots. Three consultant neurosurgeons rated it favorably on a 7-point scale, singling out the pairing of steerable navigation and direct visualization—"the catheter-based ones are blind, whereas this system allows you to navigate up the spine and visualise structures."
What to believe, and what not to yet
The mechanism is real and the multi-level evidence (force measurement, FEA, cadaver) is consistent. The load-decoupling principle is physically sound and the friction-compensation duty-cycle insight is a transferable engineering lesson. The 150 mm from standard lumbar access is genuinely a first in intact human spinal anatomy.
Now the honest limits, which the authors state plainly. The safety claim rests on one cadaver, plus two earlier specimens used for tuning. "No damage" means macroscopic visual inspection—no histology, so subclinical microstructural or vascular injury can't be excluded. No forces were measured in the cadaver; safety there is inferred from the phantom. A fresh-frozen cadaver has no CSF circulation, no physiological pressure dynamics, and degraded fluid—so real in-vivo behavior (and the lubricating effect of actual CSF, which should make things better, not worse) is untested. The FEA uses phantom material properties, not spinal tissue, and overestimates absolute forces. And it's a research prototype: no therapy-delivery channel yet, no real-time force sensing, extension length capped by hardware.
If it holds up under histology and physiological conditions, the payoff is a way to safely reach deep intrathecal targets from an ordinary lumbar puncture—turning an underused therapy corridor into an accessible one, with a camera to see where you're going.
The two sections most worth your time: the duty-cycle optimization results (Fig. 2 and the "Friction-dominated mechanics" discussion) for the counterintuitive mechanism, and the finite-element analysis (Fig. 4) for the load-redistribution story that explains why growing beats pushing.