Robots in Healthcare and Eldercare: What's Actually Deployed vs. Still a Pilot

Walk into more than 140 U.S. hospitals today and you will likely pass a robot in the hallway. It will almost certainly be a logistics robot — hauling linens, lab samples, or medication carts — not a machine helping a nurse turn a patient in bed. That distinction is the whole story of healthcare robotics in 2026: autonomous delivery and mobile-manipulator robots are genuinely, commercially deployed at scale, while robots that physically assist with bathing, dressing, lifting, or hands-on companionship care remain in research pilots or narrowly scoped state programs. Eldercare robots that talk to you are shipping by the hundreds; eldercare robots that touch you are still mostly in the lab.

What counts as “deployed” versus “pilot” in health robotics?

The healthcare-robotics field gets murky because vendors, journalists, and researchers use “deployed” loosely. For this article, deployed means a robot is running unsupervised, paid-for, revenue-generating work across many independent sites, with fleet-level operating data to show for it. Pilot means the robot exists, runs, and may even be impressive on video, but it is still confined to a handful of test sites, a university lab, or a single funded demonstration program with no clear commercial buyer yet. Judged by that bar, hospital logistics and mobile-manipulator delivery robots clear it easily. Physically assistive care robots — the ones doing hands-on caregiving tasks — do not, at least not yet.

Which hospital robots are actually running today?

Two players anchor the mature end of the market, and they operate very differently. Aethon’s TUG autonomous delivery robots are the oldest and most widespread: they operate in more than 140 U.S. hospitals and, as a fleet, complete over 50,000 deliveries per week, according to industry coverage. TUG robots move carts of supplies, meals, and specimens along fixed routes and have been a hospital-logistics fixture for years.

The newer, more capable category is the mobile manipulator — a robot that doesn’t just carry a cart but can pick up and hand over individual items. Diligent Robotics’ Moxi is the clearest example: by the time delivery-robot company Serve Robotics agreed to acquire Diligent in a deal announced January 20, 2026 and closed a week later, Moxi was operating in more than 25 U.S. hospitals and its fleet had completed over 1.25 million delivery tasks. The all-stock deal was valued at roughly $29 million, with up to an additional $5.3 million tied to earnout milestones — a sign that a physical-AI logistics platform saw enough proof in Moxi’s hospital deployment data to bet on it directly, rather than building a hospital-specific product from scratch. That acquisition logic mirrors the broader pattern of ROI-driven consolidation across physical AI deployments: once a robot’s unit economics are proven in one vertical, buyers move fast to fold it into a platform rather than compete with it.

Diligent, founded in Austin, Texas in 2017 by CEO Andrea Thomaz and CTO Vivian Chu, didn’t stop at the original Moxi. It unveiled Moxi 2.0 at NVIDIA’s GTC conference on October 28, 2025 — a next-generation mobile manipulator running on NVIDIA’s Thor compute platform, offering roughly ten times the onboard compute of the original, weighing 136 kg, and rated to carry payloads up to 15 kg. The first upgraded units were slated to start shipping to hospitals in the first half of 2026. That compute jump matters more than it might sound: it’s the difference between a robot that follows a scripted delivery route and one with enough onboard capacity to handle messier, less-structured pickup-and-handoff tasks — the kind of leap the physical AI compute stack makes possible across robotics categories, not just healthcare.

What happened when a hospital tried to scale up its robot fleet?

Deployment scale is not the same as deployment success, and healthcare robotics has one of the field’s most honest public failure cases. MultiCare Health System in Washington state began buying Moxi robots in 2023 and eventually operated 14 units across its Puget Sound-area hospitals — a real, paid, multi-site deployment by any definition. But it didn’t hold. Tacoma General removed its Moxi unit in August 2025, and by 2025-2026 the program had been suspended and the robots removed system-wide. Nurses reported that the robots needed a human escort to move between floors, which undercut the time savings the program was sold on in the first place — turning a labor-saving tool into an additional coordination burden.

That failure is instructive precisely because it wasn’t a technology failure in the narrow sense — the robots worked as designed. It was an operational failure: nobody had solved the fleet-management problem of getting robots reliably across floors and elevators without a human chaperone. It’s the same lesson that shows up across every physical-AI vertical once robots leave the demo floor and enter daily operations — deployment success depends less on the robot’s capability ceiling and more on the operational discipline behind it, the kind of day-two fleet operations work that keeps a fleet running instead of quietly getting unplugged.

Deployed or still a pilot? A category-by-category comparison

Robot category	Example	Deployment status	Scale evidence
Hospital logistics (fixed-route delivery)	Aethon TUG	Deployed, mature	140+ U.S. hospitals, 50,000+ deliveries/week
Hospital mobile manipulator	Diligent Moxi	Deployed, scaling	25+ hospitals, 1.25M+ tasks completed; Moxi 2.0 shipping H1 2026
Social companion robot for seniors	Intuition Robotics ElliQ	Deployed via state program	800–900+ units distributed free by New York State (NYSOFA) since 2022
Physically assistive caregiving robot	AIREC / Dry-AIREC	Research pilot	Lab and symposium demonstrations; Moonshot program target of 2050 for full integration
National eldercare robotics push	Japan’s Noetra consortium	Policy/investment stage, not yet deployed hardware	Target of 10 million robots by 2040 across 18 sectors; ¥380–387.3 billion committed for fiscal 2026

Are elder-companion robots actually reaching seniors, or is that still a pilot too?

Companion robots are the one eldercare category that has moved past pilot into a real, if narrowly scoped, government-funded rollout. New York State’s Office for the Aging has been distributing Intuition Robotics’ ElliQ device free to older adults since 2022. Program figures differ slightly by source and by when they were published — the ElliQ blog cites over 800 units distributed, while NYSOFA’s own February 2026 program update put the figure at over 900 — but both numbers describe the same growing, state-funded rollout, not a lab trial. NYSOFA reports that roughly 94-95% of participants showed a measurable reduction in loneliness. Importantly, ElliQ is a conversational and cognitive-engagement device, not a physical-assistance robot: it doesn’t lift, bathe, or move anyone. That’s precisely why it was deployable years before physically assistive robots were — talking is a much lower bar than touching, both technically and in terms of safety liability.

What about robots that physically help with bathing, dressing, or mobility?

This is where “deployed” turns into “pilot,” full stop. Japan’s AIREC (AI-driven Robot for Embrace and Care) program, funded through the Japan Science and Technology Agency’s Moonshot initiative and built with NVIDIA-accelerated compute, is developing robots explicitly for hands-on eldercare tasks: diaper changes, bathing assistance, and repositioning bed-bound patients to prevent pressure sores. Two variants are in active testing — Dry-AIREC, which uses NVIDIA GPU compute, and the lighter AIREC-Basic, built on an NVIDIA Jetson Orin NX module — with the latest progress shown at the International Symposium on System Integration in January 2026. But the program’s own stated horizon for full integration into eldercare settings is 2050. That is not marketing language for “coming soon” — it is a research agency’s honest timeline for a genuinely hard robotics problem: manipulating a fragile human body safely, gently, and adaptively is a different order of difficulty than picking up a supply cart in a hospital corridor.

Why is Japan betting so heavily on robots it hasn’t deployed yet?

Japan’s demographic math forces the issue regardless of where the technology stands today. In a policy update announced by METI Minister Ryosei Akazawa around June 30-July 1, 2026, the government set a target of 10 million additional robots deployed by 2040 across 18 sectors, including nursing and medical care alongside food manufacturing and other industries facing labor shortages. Implementation runs through a newly formed entity called Noetra, majority-owned by SoftBank alongside NEC, Sony Group, and Honda. For fiscal 2026 alone, METI committed between ¥380 and ¥387.3 billion to the initiative — reported by different outlets as somewhere between roughly $2.3 billion and $3.5 billion depending on the conversion used — with potential government support reported at up to ¥1 trillion (roughly $6.1 billion) over five years. That is a national industrial strategy, not a hospital procurement decision, and it underscores the gap this article is built around: the money and policy intent for eldercare robotics at national scale exist now, but the hands-on robots that would actually do the caregiving work are still years from the deployment maturity that TUG and Moxi have already reached in hospital logistics.

So where does the deployed-versus-pilot line actually sit?

Put the pieces side by side and a clear pattern emerges. Every healthcare robot that has reached genuine multi-site commercial scale — TUG, Moxi — does logistics: moving objects along known or semi-known paths inside a building. Every robot still confined to funded pilots or symposium demos — AIREC and its variants — does hands-on physical care of a human body. ElliQ sits in between: deployed at real scale through a government program, but solving a conversational and emotional-support problem rather than a physical one. That ordering isn’t a coincidence. It tracks almost exactly with how forgiving each task is of a mistake, and it’s the same ranking you’d get by looking at where physical AI has produced measurable ROI across other industries first: the tasks that get automated earliest are the ones where a failure means a late delivery, not an injury. For a deeper look at the underlying terminology this comparison relies on — mobile manipulator, autonomous mobile robot, Physical AI itself — the robotics glossary is a useful reference point. Until physically assistive care robots close that reliability gap, expect hospital hallways to keep filling with delivery robots long before nursing homes fill with robots that can safely help someone out of bed.