Service robots in hospitals: what ground robots actually do today

The starting point for any robotics project is the size and structure of the operation. Destatis counts 1,841 hospitals with 472,900 beds on annual average and around 17.5 million inpatient cases for the 2024 reporting year, at an occupancy rate of 72.0 percent and an average length of stay of 7.1 days. Every one of those cases produces transport: meals, linen, waste, sterile goods, samples, medication. That volume, not the robot, is what a project has to be dimensioned against.

There is no reliable public statistic on how many German hospitals run service robots in routine operation. Adoption percentages circulating in vendor material are not traceable to a primary source, and we do not quote them. What can be established is where the transport load comes from. Only 331 of 1,893 hospitals had their own department of laboratory medicine (Destatis hospital core data 2022), and only 91 had a department of transfusion medicine. The large majority of houses therefore move samples and blood products either across a campus or to an external laboratory, which is precisely where automated transport chains become interesting. Our healthcare vertical starts from that logistics picture rather than from a hardware catalogue.

• Routine, not care: ground robots take over repetitive haulage. They do not take over nursing tasks, and no serious project should be sold on that basis.
• Distance drives the business case: long corridors, several floors, a central sterile supply department at the other end of the site. Short single-ward routes rarely pay for themselves.
• Funding: transport, cleaning and disinfection robotics is not eligible under the German Hospital Future Act. Section 19 KHSFV provides no funding category for it. Any budget plan built on that assumption fails at the first audit.

The highest-volume application is also the least glamorous one. Heavy roll containers, food trolleys, laundry carts and waste bins move on fixed routes at fixed times, which is exactly the profile an automated guided vehicle handles well. The hardware for this is mature and specified, and it is worth reading the data sheets rather than the brochures.

• MLR Caesar Hospital II (Ludwigsburg): 500 kg payload, 1.5 m/s, LiFePO4 battery, magnetic or building-referenced navigation, stainless steel construction. MLR names hospital references including Robert-Bosch-Krankenhaus Stuttgart, Jena University Hospital and Magdeburg University Hospital.
• DS Automotion CAREY (Linz): available as spin and trike variants, 500 kg each, 1.6 m/s, SLAM or magnetic navigation, LiFePO4.
• OTSAW TransCar: 500 kg. The TransCar line moved out of Swisslog Healthcare at the end of 2021 into the joint venture OTSAW Swisslog Healthcare Robotics (OTSAW 60 percent, Swisslog 40 percent), which matters for support and spare parts questions.
• Aethon T3 / T3XL: 340 kg and 454 kg respectively, 76 cm/s, 9.0 hours on LiFePO4 with a 3.2 hour charge time, declared against CE, EN ISO 12100, EN 60204-1 and EN 60601-1-2. There is no documented DACH installation, so a German project would be building the first one.

Two things follow from this list. First, the payload classes are real and comparable, so a specification can be written before any supplier conversation starts. Second, the navigation principle is a genuine decision, not a detail: magnetic guidance is robust and cheap to troubleshoot but fixes the route into the floor, while SLAM navigation is flexible and degrades in corridors that look identical or that change layout daily. Both are represented in the robot catalogue, and the choice should follow the building, not the vendor.

Sample and medication transport is a different problem class from bulk haulage. The loads are light, the routes are frequent and irregular, and access control matters more than payload. The reference product in the DACH market is Robotise JEEVES (Munich): 45 kg payload, five individually addressable drawers, 100 litres of total volume of which up to 70 litres can be cooled to 7 degrees Celsius, up to eight hours of battery runtime, and PIN-secured compartments. Documented deployments include LMU Munich, where the robot ran inside the BMBF-funded REsPonSe project, with the transition into routine operation not publicly documented, and Prosper-Hospital Recklinghausen, where blood samples were moved from the emergency department to the central laboratory from 22 June 2024.

For lighter ward courier duty, DS Automotion SALLY offers a 100 kg base platform with a ward courier superstructure rated at 50 kg and 1.0 m/s. The distinction to draw in a specification is between a closed, access-controlled compartment and an open shelf: only the former is defensible for medication or patient-identifiable samples.

• Access control: individually addressable, lockable compartments with PIN or badge release, and a log of who opened which drawer when.
• Cold chain: if the route carries temperature-sensitive material, the cooled volume and the achievable temperature belong in the tender, together with the question of what happens when the robot is delayed.
• Cleaning and disinfection of the robot itself: no manufacturer documents a release list of approved surface disinfectants for its own housing. Ask for one in writing, and involve hospital hygiene before the pilot rather than after it.
• Sample integrity: mechanical stress on the load is a validation topic with the laboratory, not a marketing claim. Any new route should be validated against the existing manual transport.

Where the laboratory sits off site, ground transport ends at the loading bay and other modes take over. That interface between in-house robotics and external medical logistics is a separate design question. The operational logic behind staff relief is set out in the article on transport robots and clinical staff.

Floor cleaning robots and UV-C disinfection robots are frequently sold together with transport robotics, and this is where claims need the most discipline. The BHUK practice guide on scrubber driers in hygienically critical areas, dated 27 March 2024, positions cleaning robots primarily for areas without infection risk or with possible infection risk, explicitly advises against single-tank machines in hospitals, and asks operators to carry out a data protection assessment for the camera and mapping material such machines generate. Fraunhofer IPA stated in November 2022, verbatim, that no suitable cleaning robots were available on the market that could implement automated cleaning and disinfection of surfaces in a technologically and economically viable way.

UV-C is subject to physics that no product can argue away. The dose is irradiance multiplied by time and falls off sharply with distance, it acts only on directly irradiated surfaces, shadowing is the central limitation, the room has to be empty of people, and UV-C does not clean because it removes no soil. The clinical evidence is thinner than the marketing. BETR-D, the only large randomised trial (Anderson et al., Lancet 2017;389:805-814), reported a relative risk of 0.70 with a 95 percent confidence interval of 0.50 to 0.98 for the UV arm, but is internally inconsistent: the bleach plus UV arm showed the weakest and non-significant effect at 0.91. Sun et al. 2023 found an incidence rate ratio of 0.90 for C. difficile and 0.72 for VRE, neither significant, and concluded that no advantages were found for the use of UV-C in healthcare settings. Health Quality Ontario recommended against public funding in 2019. There is no Cochrane review on UV-C room disinfection and infection rates.

• Supplement, never substitute: UV-C is a documentable addition after manual wipe disinfection according to KRINKO, not a replacement for it.
• Market reality: UVD Robots units reached around 40 German hospitals through the EU emergency procurement programme of 2021 and 2022, and no controlled studies with infection rate endpoints have been published for that platform.
• An honest deployment: Hero21 (ICA Traffic, Dortmund) has been in use at St. Josef-Hospital Bochum since October 2020, described explicitly as a combination of scrub-wipe disinfection and robot.
• Regulatory framing: the KRINKO and RKI recommendations govern surface disinfection. They set the process the robot has to fit into.

The detail on dosing, shadowing and what a defensible tender text looks like is set out in the companion article on UV-C robots in clinics.

This is the point at which hospital robotics projects most often stall, and it is systematically understated in product communication. A robot that cannot change floors is a single-ward robot. Yet not one manufacturer publicly names a concrete elevator protocol or an elevator vendor partnership. No KONE, no Schindler PORT, no Otis, no thyssenkrupp interface is documented as a product feature. The two partial exceptions are Relay Robotics, which names elevator brands, and the Pudu CC1 Pro, which lists elevator control as an option. Fire door control is documented by nobody at all.

The practical consequence is that elevator and door connectivity is an integration item in every clinical project, not an off-the-shelf product feature. It is scoped, engineered, tested and accepted separately, usually together with the elevator service provider, building automation and fire protection. Treating it as included in the robot price is the single most common source of budget overrun. This is exactly the work a manufacturer-independent integrator exists to do, and it belongs in the business case from day one, which is why our ROI calculator treats it as a separate cost line rather than folding it into the hardware.

• Elevator interface: clarify early whether the control system is addressed via dry contacts, a PLC, or a manufacturer interface, who owns the elevator maintenance contract, and whether that contract permits third-party connection at all.
• Priority and mixed use: whether the robot gets its own car, a priority call, or shares with patient transport is a hygiene and workflow decision as much as a technical one.
• Doors: automatic door drives can usually be triggered. Fire section doors must fail safe and are subject to fire protection approval, so they are rarely a simple integration.
• Network coverage: elevator shafts and stairwells are the classic coverage gaps. Measure them before, not after, the pilot.
• Acceptance: define the interface as a separate deliverable with its own test cases, so that a failed elevator call is attributable and not a dispute between three suppliers.

The regulatory framing for hospital ground robots is more settled than the market suggests, and getting it right removes a lot of noise from a tender.

ISO 3691-4 is the relevant standard for driverless industrial trucks and transport robots. The updated version dates from March 2024 and has been listed as harmonised in the EU Official Journal since May 2024. In practice, almost no manufacturer publishes a declaration of conformity against it, which is precisely why it should be a named requirement in every tender document rather than an assumption. ISO 13482 does not apply here: it covers personal care robots, not transport robots operating in a hospital corridor, and quoting it as the safety basis for an automated guided vehicle is a category error that appears regularly in vendor material.

The second distinction is between machinery law and medical device law. Under Regulation (EU) 2023/1230, the Machinery Regulation, which applies from 20 January 2027, a robot that transports meals, linen, waste or sealed samples is a machine. Regulation (EU) 2017/745, the MDR, only applies once the manufacturer assigns a medical purpose to the device, for example dosing or administering a medicinal product. That single question, has the manufacturer declared a medical intended purpose, determines the entire conformity route, the documentation burden and the responsibilities of the operator.

• Ask for the declaration of conformity and the applied standards in writing, with ISO 3691-4 named explicitly.
• Clarify whether the manufacturer claims a medical intended purpose. If not, MDR obligations do not attach, and nobody should imply otherwise.
• For cleaning and disinfection, the KRINKO and RKI recommendations define the process the machine has to support.
• Note that the operator carries obligations of its own under occupational safety and workplace law once the machine shares corridors with staff and patients.

Hospitals rarely need one robot class. A campus that runs bulk transport, ward courier service and floor cleaning is looking at two or three hardware families, and in most cases at more than one manufacturer. That is manageable, provided the interfaces are specified before the purchase rather than after it.

VDA 5050 is the open communication interface between driverless transport vehicles and a master control system. It is not a cure-all, and implementations differ in depth, but it is the realistic basis for operating heterogeneous fleets from one control layer instead of one dashboard per vendor. Where a manufacturer does not support it, that should be a scored criterion in the evaluation, not a surprise during commissioning.

• Specify the workflow, not the product: routes, volumes, time windows, load carriers, floors and access rights. The hardware shortlist follows from that and can then be compared on equal terms.
• Interfaces as deliverables: elevator, doors, network, fleet control and any connection to hospital IT belong in the scope with their own acceptance criteria.
• Service footprint: response times, spare part availability and technician coverage in the region. A platform without a documented DACH installation is a different risk profile from one with named hospital references.
• Exit: data ownership for maps and logs, and whether a replacement of the fleet management layer is possible without replacing the vehicles.

werob is a manufacturer-independent systems integrator for service robotics and a brand of CITO GmbH, Hamburg. We do not build robots and we do not sell a single manufacturer's line. The value is in specifying the operational case correctly, in comparing hardware against it honestly, and in owning the integration work that the product data sheets leave out.