Smart robots in healthcare: 5 ways they’re helping patients and doctors

Surgeons performed roughly 3,153,000 procedures with da Vinci systems in 2025, according to Intuitive’s preliminary full-year results, about 18% more than in 2024. The same release counts 870 da Vinci 5 placements for the year, up from 362. Those numbers say something the robot hype cycle keeps missing. The robots doing the most good in hospitals today are mostly not the humanoids in the headlines. They’re specialized machines that take over precise, repetitive or risky work, and they’ve been doing it quietly for years.

I call them boring robots. An instrument arm that doesn’t tremble at hour four of a prostatectomy. A compounding cell that weighs every syringe before it lets a chemo bag leave the hood. A cart that rides the elevator to the fourth floor at 2 a.m. with IV antibiotics in a locked drawer. None of them will walk onstage at a keynote. That’s rather the point.

The sales figures agree. In its World Robotics 2025 service robots release, the International Federation of Robotics counted around 16,700 medical robots sold in 2024, up 91%. Demand for surgery robots rose 41%, and rehabilitation and non-invasive therapy robots more than doubled. That’s a lot of hardware heading into buildings that were designed for people, gurneys, and paper.

What follows are five robot types earning their keep right now, written as spec cards. Same four lines for each: what it does, where it works, what the evidence actually says, and where it falls short.

1. Surgical robots extend a surgeon’s hands, not their judgment

Surgical robots are the most mature category and by far the most expensive. The surgeon sits at a console, the system scales and filters hand motion into wristed instrument movement, and the patient gets smaller incisions. Intuitive dominates soft-tissue surgery, Stryker’s Mako has a firm hold on joint replacement, and CMR Surgical’s Versius and Medtronic’s Hugo compete in the markets where they’re cleared.

• What it does: Gives the surgeon 3D high-definition vision and articulated instruments for minimally invasive work in urology, gynecology, general and thoracic surgery. The da Vinci 5, cleared by the FDA in March 2024, adds force feedback so the surgeon can feel tissue resistance through the controls.
• Where it works today: Most large US hospitals, plus a growing share of community hospitals. The more-than-doubled da Vinci 5 placement count in 2025 suggests replacement cycles are already underway at sites that bought earlier generations.
• What the evidence says: Adoption has outrun proof. A JAMA Network Open study of 73 Michigan hospitals found the robotic share of common general surgery operations climbed from 1.8% to about 15% between 2012 and 2018, and its authors warned that uptake was outpacing the evidence on safety and cost. For many routine procedures, the business case gets made on surgeon ergonomics and conversion rates rather than on dramatic clinical gains over laparoscopy.
• Limits: Capital cost, per-case instrument costs and a real learning curve. Robot time also competes for OR time, which means scheduling becomes part of the purchase whether anyone planned for it or not.

A surgical robot lives inside one room with one team. The other four on this list have to move through the whole building, and that’s where I’ve seen projects stall. I’ve worked on healthcare software and automation projects for years, and the pattern repeats. The robot is rarely the hard part. The hard part is the Wi-Fi dead zone outside radiology, the elevator controller nobody has a manual for, the pharmacy system that exports orders in a format someone chose in 2009. A robot that can’t see the order queue is a very expensive statue.

So the software layer matters as much as the motors. In the deployments I’ve watched succeed, AI agents in healthcare that coordinate scheduling, tasks, and data between the EHR, the pharmacy and the robot fleet were what let physical machines fit into hospital workflows instead of running beside them. Get that layer wrong, and you get a lovely pilot. And nothing after it.

2. Pharmacy robots beat human accuracy, not always human speed

Chemotherapy compounding is where pharmacy automation makes its clearest case. Every dose is patient-specific. The drugs are hazardous to the person mixing them. A dosing error isn’t a typo, it’s a clinical event. Systems like APOTECAchemo and Kiro Oncology do the vial and syringe work inside a sealed cell, check each step by weight or image, and log the result. Down the hall, central-pharmacy robots and carousels handle the duller job of picking and packaging unit doses.

• What it does: Compounds IV doses, especially cytotoxic ones, and picks, packages and labels oral medications for distribution. The goal is fewer hands on hazardous drugs and a full audit trail for every bag.
• Where it works today: Mainly oncology pharmacies in large academic and cancer centers, plus central pharmacies that supply several hospitals in a system.
• What the evidence says: A 2023 systematic review and meta-analysis in Medicina covering 14 studies found robotic compounding more accurate than manual work, with failure rates from 0.9% to 16.75% against a 5% accuracy threshold. Infusion bags prepared by robots had lower contamination (odds ratio 0.176). Speed was a split decision. Two studies found robots slower, two found them faster.
• Limits: Throughput depends heavily on drug mix and model, and the same review found no significant advantage for robots on contamination of the compounding area itself. Pharmacists still verify output, so the staffing gain is smaller than the sales deck implies.

One honest note from the projects I’ve seen. The robot usually wins on the hardest, most dangerous preparations and loses on simple high-volume ones that a skilled tech can do faster by hand. Plan the split before you buy, not after.

3. Delivery robots give nurses back their walking time

Hospital logistics is the purest boring-robot job there is. Somebody has to move meds, specimens, linens, meals and supplies across a building all day, and in plenty of hospitals that somebody is a nurse or tech who should be doing something else. Aethon’s TUG has hauled carts with locked drawers and elevator access for years. Diligent Robotics’ Moxi adds an arm for fetch-and-deliver tasks. Diligent markets Moxi as a humanoid. It’s a one-armed cart with a friendly face, which I mean as a compliment.

• What it does: Carries medications, lab specimens, supplies and linens between departments on its own, including elevator rides and drop-offs to secure drawers, triggered by staff requests or by the pharmacy system.
• Where it works today: Diligent reported in February 2025 that its Moxi fleet had passed 1 million deliveries across 23 health systems and 31 hospital-level partnerships, saving clinical staff more than 575,000 hours.
• What the evidence says: Mostly vendor-reported and operational, not peer-reviewed. Spread evenly across 31 hospitals, 575,000 hours works out to roughly 18,500 hours per site, which is close to nine full-time positions’ worth of walking. That’s my arithmetic, not Diligent’s, and real distribution across sites is surely lumpier.
• Limits: Elevators, automatic doors, Wi-Fi coverage and badge access all need integration work before the first run. Crowded corridors slow robots down. And staff stop trusting a delivery robot remarkably fast after one missing specimen.

4. UV-C disinfection robots help, but only after the cleaning crew is done

UV-C robots roll into an empty room after discharge and bathe the surfaces in germicidal light. They got famous during COVID-19, when every hospital wanted to be seen doing something visible about infection. The evidence behind them is better than for most hospital robots. It’s also more mixed than the marketing.

• What it does: Delivers ultraviolet light to surfaces in patient rooms and ORs after manual cleaning, as an extra kill step against organisms like MRSA, VRE and C. difficile.
• Where it works today: Terminal cleaning of isolation rooms, operating rooms between cases and high-risk units such as oncology and transplant.
• What the evidence says: The BETR-D trial, published in The Lancet in 2017, ran in nine hospitals in the southeastern US and included 21,395 patients. Adding UV to standard quaternary ammonium cleaning cut acquisition of target organisms from 51.3 to 33.9 cases per 10,000 exposure days, a relative risk of 0.70. Adding UV to bleach did nothing measurable for C. difficile: 30.4 versus 31.6 cases.
• Limits: UV works by line of sight, so shadowed spots under bed rails and inside drawers get a weaker dose. The room has to be empty, which adds minutes to turnover at exactly the moment bed management wants it back. It supplements manual cleaning. It doesn’t replace it.

5. Exoskeletons get stroke patients walking more often than therapy alone

Rehabilitation robots range from treadmill-based gait trainers to wearable exoskeletons like Ekso Bionics’ EksoNR and ReWalk, now sold by Lifeward. Therapists use them to fit far more steps into a session while taking the lifting strain off their own backs. Of the five categories here, this one has the evidence that would hold up best in front of a skeptical clinical committee.

• What it does: Supports and powers hip and knee motion so patients after stroke, spinal cord injury, or other neurological damage can practice walking with less hands-on help from staff.
• Where it works today: Inpatient rehab hospitals and outpatient neuro clinics, with personal devices for home use in some spinal cord injury cases. ReWalk was the first wearable motorized exoskeleton cleared by the FDA, in June 2014, for use with a trained companion. EksoNR now carries FDA clearances for stroke, spinal cord injury, acquired brain injury, and multiple sclerosis rehab.
• What the evidence says: A 2020 Cochrane review of 62 trials with 2,440 participants found that electromechanical and robot-assisted gait training plus physiotherapy raised the odds of walking independently after stroke (odds ratio 2.14), with high-certainty evidence. Of 100 patients who can’t walk on their own, about 62 would regain independent walking with device-assisted training, versus 45 with physiotherapy alone. The review pools gait trainers and exoskeletons, so don’t read it as proof for one brand.
• Limits: Fitting and setup eat session time, and therapists need real training before they use the devices confidently. Personal exoskeletons for home use remain expensive, and reimbursement is still settling.

Robots won’t replace nurses. The workforce math points the other way.

Give it ten years and robots will replace most of the nursing staff.

That’s the line you hear at every health tech conference, usually from someone who has never watched a nurse do a med pass on a full unit. The workforce projections say the opposite problem is the real one. HRSA’s December 2025 nursing projections show a projected 8% shortage of registered nurses in 2028, easing to 3% by 2038, which is still a gap of 108,960 full-time equivalent RNs. For licensed practical nurses, it gets worse over time. HRSA expects LPN supply to meet just 70% of demand in 2038, a shortfall of 245,950 FTEs.

Look back at the five spec cards. Not one of those machines does nursing. They do the walking, the mixing, the wiping and the lifting that surround nursing. That isn’t replacement. It’s a way of buying back hours from a workforce hospitals already can’t hire enough of.

The robots that matter most over the next decade will be the ones that hand those hours back to bedside staff without creating new chores, like babysitting a stuck cart at the elevator. HRSA projects the shortage. The conclusion about robots is my own.

The evidence behind each robot is uneven, and buyers should price that in

Vendors present every category as proven. It isn’t. Here’s how the five compare when you strip out the brochure language.

Notice the last column. Three of five barriers are workflow, not price. In my experience the purchase order is the easy meeting. The hard meetings come later, with facilities, IT security, infection control and the night shift charge nurse who has to live with the thing.

Questions a hospital should ask before buying a robot

1. What task does it take off a person’s plate, in hours per week? If the vendor can’t name the task and estimate the hours for your units, keep the checkbook closed.
2. Which systems does it need to talk to? EHR, pharmacy, elevators, doors, nurse call and badge access each count as a separate integration project.
3. What does the evidence show for our case mix? Ask for data from sites that look like yours, not the vendor’s best reference hospital.
4. Who owns it at 3 a.m.? Someone has to clear a jam, reboot a frozen unit or rescue a cart stuck by the loading dock.
5. How long until staff trust it? Budget training time, and decide in advance how you’ll recover adoption after the first failed delivery or missed dose.
6. What’s the full cost per use? Count instruments, consumables, service contracts, software licenses and the room or OR time the robot occupies.
7. How will we know in six months whether it worked? Pick one metric before go-live and agree on what number means keep, fix or return.

The best hospital robot is the one nobody bothers to mention by its second month on the floor.