a tertiary hospital in China, six-month single-centre study, anonymized
Ten Delivery Robots Outpaced 19 Manual Workers in a Hospital Study
A six-month peer-reviewed study of 10 hospital delivery robots versus 19 manual workers: 32 to 36% faster deliveries and 100% verification accuracy.
- 32 to 36%
- Shorter delivery time than manual runs
- 7.3x
- More trips: 10 robots vs 19 workers
- 100%
- Verification accuracy in robot group
- 6.928M RMB
- Modeled 10-year savings, per authors
Based on a documented real-world deployment. Figures are from public reporting; the organization is not named.
Drug and specimen runs that ate staff time
Inside a tertiary hospital in China, medications and laboratory specimens moved by hand. Staff carried drugs from the pharmacy to the wards and specimens from the wards to the laboratory, trip after trip, through a building where every run competed with clinical work.
The six-month, single-centre study set out to measure what that manual model cost in time and reliability. The baseline figures were clear enough to justify a controlled trial, and they set the bar any robot fleet had to clear.
Speed was only part of the problem. A delayed or misverified item lands on a nurse's desk and a patient's chart, so reliability carried as much weight as minutes.
- Pharmacy to ward, manual average: 31 minutes (plus or minus 6.2)
- Ward to laboratory, manual average: 38 minutes (plus or minus 7.2)
- Manual verification accuracy: 97%
Ten robots, one parallel-controlled comparison
The study fielded 10 autonomous mobile robots and ran them against a manual group of 19 workers over six months, in a parallel-controlled design. That structure lets the authors compare two live operations side by side, rather than a before-and-after snapshot that shifting staffing or case mix could distort.
Routing and coordination were the engineering core. Dijkstra's algorithm handled global path planning, while A* combined with the Dynamic Window Approach handled local navigation around people and obstacles. An M/M/c queuing model coordinated the robots as a group, so jobs were assigned across the fleet rather than each unit working its own queue.
A deployment like this carries several layers a hospital must plan for: route design, queue policy, and the handoffs at each ward and laboratory.
- Map the pharmacy, ward and laboratory routes and set global paths
- Configure local navigation so robots avoid people and objects in corridors
- Coordinate the 10 robots with a queuing model across the whole fleet
- Run the robot group and the manual group side by side for six months
- Measure per-route times, verification accuracy, omissions, and staff and patient satisfaction
What the six months measured
Delivery time fell 32% to 36% against manual delivery, with p below 0.001. Per route, the pharmacy-to-ward run dropped from 31 minutes to 21 minutes, and the ward-to-laboratory run dropped from 38 minutes to 23 minutes. Both route comparisons were significant at p below 0.001.
Throughput is the number that changes staffing conversations. The 10-robot fleet completed 7.3 times more delivery trips than the 19 manual workers over the same six months. That ratio compares a fleet with a team, so it describes capacity, not a one-for-one swap of individuals.
Verification accuracy reached 100% in the robot group, against 97% for manual delivery. The robot group also recorded a lower omission rate (p = 0.011), and nurse and patient satisfaction were both significantly higher in that group (p below 0.05).
On cost, the authors report that over a 10-year lifecycle the robot system saved 6.928 million RMB, and that sensitivity analysis supported the economic result. Treat that as the model's output rather than a booked saving, and test it against your own labor costs and fleet terms.
The limits matter. This was one hospital and a drug and specimen scope. A 2025 prospective study at a different tertiary hospital, covering 122 non-urgent medication missions, found elevator congestion strongly associated with delivery failures, with most failures caused by passengers or cargo physically obstructing the robot. A 2024 qualitative study of cobots in a 900-bed hospital found they often needed human support to finish deliveries within 30 to 60 minutes across multiple floors.
Where a fleet like this fits a lease
Service Robot Co. is a full-service, vendor-neutral commercial robot integrator for US businesses. We select robots across manufacturers, then finance, deploy, integrate, train and service every unit through a nationwide US engineer network. For a hospital-style delivery fleet, the robot is the smaller share of the work. Route mapping, elevator and door integration, staff training and ongoing service decide whether the figures in a study like this one show up on your floor.
Our offer is a lease, not a sale. We lease robots on a 3-year lease with one fixed payment, quoted after a site assessment. Because we are not tied to one manufacturer, we can compare platforms against your routes, floors and load profile before recommending a fleet, including a fleet that a site may not yet be ready to support.
Healthcare buyers usually ask about accuracy and safety before cost. This study gives a peer-reviewed baseline for both. Our job is to test that baseline against your building, with go-live support and a service plan that includes remote triage and on-site dispatch when a unit goes down.