Robo2u
All posts

The Robot Dental Assistant Nobody Is Building

Dental robots automate the dentist's hands while the assistant's role stays entirely human. Why suction is the harder engineering problem.

By Robo2u Editorial · 18 min read

Every dental robot that has reached a patient was built to replace the dentist. Yomi guides an implant drill. Perceptive prepped a crown. Both aim at the person holding the handpiece, because that is where the precision lives and where the billing lives. Meanwhile the second person in the room, the one running suction, holding retraction, passing instruments and reading the operator's next move before it happens, has no robot pointed at them at all.

That is backwards from what an engineer would predict. Drilling a tooth is a millimetre-scale positioning task on a rigid target with a pre-operative 3D plan, which is exactly the kind of problem robots have solved for thirty years in surgery and manufacturing. Assisting is a continuous, two-handed, anticipatory task in a wet cavity that moves, on a patient who moves, with no plan beyond what the operator decides in the next two seconds. The first is a robotics problem. The second is closer to a general manipulation problem, and general manipulation is not solved.

This piece works through what a dental surgery assistant actually does, why each part of it resists automation, what the real dental robots do and do not do, where the "world's first fully automated dental procedure" headline gets misreported, and what voice control has genuinely delivered since the early 2000s.

The take: Dental robotics has automated the precision task and left the dexterity task alone. Yomi is a haptically constrained implant guide that still needs a surgeon driving it, and Perceptive's autonomous crown prep has no FDA clearance despite widespread reporting that it does. The dental surgery assistant remains unautomated for a hard engineering reason: high-volume evacuation in a deformable, occluded, wet field with a moving patient is a tougher manipulation problem than drilling a planned osteotomy. Voice control has quietly worked for two decades, and its reach stops at the equipment and the record.

Companion reading: surgical & medical robots, robot teleoperation, collaborative robots, end-effectors & grippers, force-torque sensing, and robot safety & functional safety.

Table of contents

  1. What a dental surgery assistant actually does
  2. Four-handed dentistry as a control problem
  3. The real dental robots, and what they do
  4. The autonomous crown headline, corrected
  5. Why suction is harder than drilling
  6. What voice control actually automates
  7. What a real assistive system would look like
  8. Limits and honest reality

What a dental surgery assistant actually does

Ask someone outside dentistry what a dental assistant does and you will hear "passes instruments". That is a fraction of it. Across a routine restorative appointment the assistant is running several concurrent jobs:

  • High-volume evacuation. Positioning a suction tip to clear water, saliva, blood and debris from a field that is being flooded by the handpiece spray, without blocking the operator's view, without touching soft tissue hard enough to hurt, and without sucking the tongue or cheek into the tip.
  • Retraction. Holding cheek, tongue and lip clear of the working field, adjusting continuously as the operator changes angle.
  • Moisture control. Keeping the tooth dry enough that a bonded restoration will actually bond. This is the difference between a filling that lasts a decade and one that fails in a year.
  • Instrument transfer. Delivering the right instrument into the operator's hand in the right orientation, usually without being asked, then taking back the used one.
  • Material handling. Mixing and loading materials with working times measured in tens of seconds.
  • Patient management. Reading distress, warning before the uncomfortable part, adjusting the chair, keeping a nervous patient in the chair at all.
  • Infection control and charting. Barrier technique, instrument cycles, and recording what the operator calls out.

Only one of those, instrument transfer, resembles a classical pick-and-place task. The rest are continuous-contact manipulation in a confined, deformable, occluded space, interleaved with social judgment.

War story: The single most common thing a new assistant gets wrong is not instrument selection. It is parking the suction tip where the operator wants to look. The tip has to be close enough to clear the field and never in the optical path between the operator's eyes and the tooth, and that path changes every few seconds as the operator moves. Nobody writes that down. It is learned by being told to move it, repeatedly, for weeks.

Four-handed dentistry as a control problem

"Four-handed dentistry" is the standard operating model: operator and assistant working as one unit around a seated patient, with defined zones and a shared expectation of who reaches where. Stated as a control problem, it has properties that are individually manageable and jointly brutal:

Property Why it is hard
Shared workspace, no collision margin Two pairs of hands inside a cavity a few centimetres across. There is no safe separation distance to enforce, which is how most collaborative robot safety is normally achieved
Deformable, self-occluding environment Tongue, cheek and lip move, deform under contact, and hide the target. Vision is blocked by the very tissue being retracted
Wet and specular Water spray, saliva, blood. Optical sensing degrades exactly when the field is busiest
Non-stationary target The patient breathes, swallows, flinches, and sometimes moves without warning
Intent is implicit The assistant acts on the operator's next move, inferred from posture and instrument angle, not on a command
Contact is the point Retraction is sustained force against living tissue. There is no "avoid contact" fallback

Compare that with an implant osteotomy: a rigid mandible, a CT-derived plan, a target that does not deform, and a task defined before the patient sits down. Robotics solved the second kind of problem. It has not solved the first.

The real dental robots, and what they do

Two systems dominate the conversation, and neither is an assistant.

Yomi, by Neocis. A haptically constrained robotic guidance system for dental implant surgery. It holds and constrains the handpiece so the drill cannot leave the planned trajectory, in the same architecture family as orthopedic systems like Stryker's Mako. The surgeon still drives. Its regulatory history is the clearest signal of what it is:

Year FDA clearance
2016 Dental implant procedures
2020 Full-arch implant treatment
2022 Bone reduction

Neocis and Yomi S were named winners of the 2026 TAG Awards for MedTech in Robotic & Procedural Innovation. This is a mature, cleared, commercially deployed system, and it automates constraint rather than action.

Perceptive. A Boston company building genuinely autonomous restorative dentistry, combining an intraoral optical coherence tomography scanner with an AI-planned robotic arm. In July 2024 it reported the first fully automated procedure on a human patient, a crown preparation completed in roughly 15 minutes against a claim of being several times faster than a human operator.

Note what both have in common. Yomi constrains a drill. Perceptive drives a drill. Neither one holds a suction tip.

The autonomous crown headline, corrected

This is the part most coverage gets wrong, and it is worth stating precisely because it keeps being repeated.

Multiple outlets reported that Perceptive's dental robot had been "approved by the FDA". It has not been. Perceptive's own site states that there has been no US-based testing under IRB approval, no Investigational Device Exemption approved by the FDA, and that the prototypes do not hold 510(k) marketing clearance. The 2024 procedure was not performed under US regulatory oversight, and FDA authorisation is the precondition for the larger regulator-reviewed testing the company says it wants to run.

Rule of thumb: When a medical robotics headline says "first" and "FDA approved" in the same breath, check the manufacturer's own regulatory statement before repeating it. Companies are legally obliged to be accurate on their own site in a way that a news aggregator is not.

None of this makes the demonstration unimpressive. An autonomous system that plans and executes a crown prep on a live human is a real milestone. It is a milestone at the prototype and pre-clearance stage, which is a different thing from a device you will meet in a surgery, and the distance between those two states is usually measured in years.

Why suction is harder than drilling

Take the single most mundane assistant task and specify it as an engineering requirement. A high-volume evacuator tip must:

  1. Track a target region defined relative to a tooth that is itself being modified in real time.
  2. Maintain a standoff close enough to capture aerosol and pooled fluid, without contacting sensitive soft tissue at pressure.
  3. Never enter the line of sight between operator and working field, where that line is not measured and changes continuously.
  4. Avoid trapping tongue or cheek mucosa against the tip, which is painful and can bruise.
  5. Respond to sudden patient movement with compliance rather than resistance.
  6. Do all of this in water spray that degrades optical sensing, and in a cavity that self-occludes.

Requirements 3 and 6 are the killers. Requirement 3 needs a model of the operator's visual attention, which the system cannot measure without instrumenting the operator's head and eyes. Requirement 6 breaks the sensing modality most manipulation stacks depend on, exactly when the task is most demanding. Force-torque sensing helps with the contact requirements and does nothing for the occlusion problem.

The drilling task, by contrast, has a pre-operative plan, a rigid registration target, and a defined stopping condition. That is why it went first. It was simply the tractable one.

What voice control actually automates

Voice-activated dentistry is often discussed as though it is arriving. It arrived a long time ago, and it is worth being clear about what it does.

Speech-recognition control of dental operatory equipment goes back to patents granted in the early 2000s, including systems for issuing spoken commands to position the dental chair and light and to drive X-ray system components, plus voice-driven selection of patient records. Current commercial products in this line include hands-free practice-management control and dictation systems.

The genuinely useful application is voice-driven periodontal charting. Calling out pocket depths, bleeding points, recession and furcation hands-free removes the classic two-person bottleneck where one person probes and another types, with reported savings above 20 minutes per hygiene visit. Hands-free chair, light and imaging control also has a real infection-control rationale, since it lets the operator adjust equipment without breaking sterile technique.

But notice the category. Every one of these automates the equipment or the record. Voice control removes the need for someone to touch a keyboard, and it leaves the tongue-retraction problem exactly where it found it. The tasks voice control has eliminated are the ones that were already discrete, verbal and low-dexterity, which is to say the ones least like the core of the job.

What a real assistive system would look like

If someone did build toward this, the sequence that follows from the analysis above is incremental rather than heroic:

Stage 1, static assist. A passive positionable arm holding a retractor or a saliva ejector at a place the operator sets by hand. No autonomy, no sensing. This exists in adjacent forms already and is the least interesting and most immediately useful step.

Stage 2, compliant held-position assist. The same arm with force-torque sensing and active compliance, so sustained tissue contact is pressure-limited and sudden patient movement produces yielding rather than injury. Still commanded, not autonomous.

Stage 3, cued repositioning. Voice or foot-pedal commands to move the tip between a small set of named positions. This is the point where the existing, proven voice-control stack becomes genuinely useful for manipulation rather than for equipment.

Stage 4, anticipatory assist. Inferring the operator's next move from instrument pose and posture. This is the research frontier and the first stage that requires solving the intent problem.

The commercial logic explains why nobody is climbing this ladder. Stage 1 competes with a person who also mixes materials, manages the patient and runs sterilisation. The robot would have to be very cheap to beat a human who does eight jobs, and dental practices are small businesses with tight capital budgets. The implant robot, meanwhile, sells against a high-value billable procedure.

Limits and honest reality

A few things this piece does not claim.

No robot replaces a dental surgery assistant today, and none is close. There is no cleared device, no prototype demonstration, and as far as published work goes, no serious programme aimed at the full role.

The precision robots are real and are working. Yomi has held FDA clearance since 2016 and has expanded that clearance twice. Dismissing dental robotics as hype gets it as wrong as the "FDA approved" headlines do, in the other direction.

Autonomy in restorative dentistry is genuinely coming, slowly. Perceptive's demonstration was real. The binding constraint is the clearance pathway rather than the science.

Voice control is mature and underused. The perio-charting case in particular is available now and has a measurable time saving, which makes it the most concrete thing in this entire article for a practice to act on.

The interesting conclusion is the inversion. In most industries automation takes the low-skill support role first and leaves the expert alone. Dentistry has gone the other way, automating the expert's most precise motion while the support role remains entirely human. The reason is that assisting is a general manipulation problem in a wet, deformable, occluded, moving workspace, and that is the problem robotics has not solved anywhere.

Sources worth reading

Related guides