Handing veneer tooth prep to a robotic arm: first ask who's accountable

At last month's dental expo, I stopped at the Lupin Dental booth for quite a while. I'd originally come to look at something else, but the line "minimally invasive tooth preparation for aesthetic veneers" caught my attention.

Tooth prep is a job a friend who runs a clinic has complained to me about. For a veneer to sit flat, you first have to evenly remove a thin layer of the tooth surface—reportedly on the order of a few tenths of a millimeter. Grind too much and the porcelain can't cover it; grind too little and the veneer protrudes, throwing off the bite. That little bit of margin is all down to years of experience.

Static guides are the old approach. A resin replica is fitted over the tooth, and the drill goes through pre-drilled holes, with the direction locked in. Cheap, one-shot molding—but once the position is set it can't be changed, and when the mouth opening is small in the posterior region, it basically can't reach in.

The robotic route: pre-op intraoral scanning and modeling, plan the removal amount, and the robotic arm follows the path. When I was going through the literature I saw two key components—one is the robotic arm providing physical guidance with force feedback during positioning, the other is a patient tracker attached to the tooth so the system can compensate when the person moves. I'm familiar with this logic; the AGVs in our warehouse work the same way—the positioning reference has to follow the thing that moves, it can't be nailed to the floor.

I couldn't actually grind a patient's tooth, so I asked the engineer to run the full workflow on an extracted-tooth model at the demo station: intraoral scan, plan removal amount, registration, robotic arm execution.

The registration step is aligning the scanned model with the robotic arm's coordinates; the engineer manually clicked a few points, and it took a fair amount of time. On screen, a planned path was overlaid with an actual trajectory, with deviations marked in color. I watched for a while—the green basically never turned yellow.

Once running, the path was very stable. I reached out and gave the end effector a light push, and there was obvious resistance and rebound—not a hard stop, more like pushing against a spring. This was better than I expected; it shows the force control isn't just for show.

The sticking points are registration and tracking. The tracker has to be fixed to the dentition, and once there's a lot of saliva or the mouth opening isn't wide enough, the field of view breaks. The engineer himself admitted the posterior region is the most troublesome. I wrote a line last month: the boundary of full automation is how long you can go without checking on it before something goes wrong. This thing still needs someone watching the whole time, ready to take over if there's a problem.

Where it's solid is very concrete: the removal amount is controllable, the path is reproducible—do the same case a second time and you get the same result; pre-op planning can be archived, and if something goes wrong you can replay and review it, which beats hand feel; the force-feedback resistance feels real, not pure position control. Where it falls short is equally concrete: the registration step is time-consuming, and any patient movement means re-aligning; the tracker is vulnerable to occlusion and saliva, and the posterior region is a hard flaw; single-unit throughput is limited by the pace of the whole digital workflow—whichever link is slow, whether intraoral scan, design, or fabrication, it just waits there; and it needs someone who knows how to maintain it—it's not plug-and-play out of the box.

So this thing isn't suited to a newly opened small clinic, nor to a place that does only a handful of aesthetic restorations a year—the equipment depreciation can't be spread out. Its users are chains or prosthodontic specialty practices that already have an intraoral scan plus CAD workflow running, a stable veneer volume, and dedicated digital staff. You have to run the numbers yourself based on case volume; I can't work that out from here.

Buying the machine saves the doctor's hands, but it doesn't save the attention needed to maintain that machine.