cs.ROOct 5, 2026
SaveA State Based Dispatch Controller for Hospital Delivery Robots with Shared Human and Infrastructure Resources
Organizations: Faculty of Electrical Engineering Warsaw University of Technology Koszykowa 75, 00-662 Warszawa · Faculty of Control, Robotics and Electrical Engineering Poznań University of Technology Piotrowo 3a, 61-138 Poznan
Abstract
Robot delivery studies can overstate transport capacity when travel to pickups and human support fall outside the modeled schedule. We formulate a location aware dispatch model that couples robot admission to transporter support and shared elevators, charging, and cleaning. The model replays 33,079 observed hospital requests; missing contents, deadlines, staffing, and completion times remain explicit scenario assumptions. Two full grids compare human dispatch, a resource aware controller, and a proximity and workload benchmark in 30 paired replications per scenario. An exploratory extension tests a simpler deadline admission rule under the same operating model. Pickup travel increases demand on a pooled elevator bank. In one high staffing case, raising assumed elevator capacity from two to four reduces human only lateness from 55.11% to 3.18%, exceeding the dispatch differences. Direct policy comparisons and robot coverage distinguish admission selectivity from system service. The analysis explains why a robot can pass a deadline admission test yet delay service through staff handoffs and shared facilities. Its contribution is a reproducible evaluation of coupled dispatch workflows, with a clear separation between admission, completion, and return to availability. The results are conditional comparisons, not estimates of hospital benefit or released clinical capacity.
Figures & tables
Fig. 1 : Illustrative hospital delivery robot installations. The depicted platforms were not used to calibrate or validate the simulation.
| Class | Share (%) | Priority | Deadline (min) | Loading (min) | Receiving (min) |
|---|---|---|---|---|---|
| Medication | 14.958 | 2 | 35 | 1.4 | 1.0 |
| Specimen | 84.061 | 1 | 25 | 1.0 | 0.8 |
| Supply | 0.929 | 4 | 120 | 2.0 | 1.5 |
| Instrument | 0.053 | 2 | 45 | 2.5 | 1.5 |
TABLE I: Synthetic task assumptions. Priority 1 defines urgent requests. Handling times are fixed inputs, not transport observations.
Fig. 2 : System architecture and decision loop. ARAD observes queue, staff, fleet, and infrastructure state when a request arrives. The controller selects a resource class. Robot execution can generate new human support work and contention for shared infrastructure, which changes the state observed at later decision epochs.
| Parameter | Reference value | Interpretation |
|---|---|---|
| Hospital beds | 200 | Normalization denominator |
| Warmup period | 7 days | Excluded from outcome measurement |
| Measurement period | 30 days | Included in every principal run |
| Night staffing | 2 transporters | 22:00 to 06:00 |
| Day staffing | 8 transporters | 06:00 to 14:00 |
| Evening staffing | 7 transporters | 14:00 to 22:00 |
TABLE II: Reference model inputs. Values are assumptions unless identified as empirical inputs in the text.
| Experiment | Simulations | Principal purpose |
|---|---|---|
| Staffing calibration | 5,040 | Select shift counts meeting service and utilization targets |
| Robot policy comparison | 8,100 | Compare human, overflow, and priority policies |
| Global sensitivity | 4,096 | Vary nine uncertain inputs simultaneously |
| Elevator contention | 2,400 | Test shared elevator capacity and service time |
| Battery and charging | 2,700 | Test energy, charger count, and charger power |
| Exact route batching | 1,400 | Test collection windows and batch capacity |
TABLE III: Experiment inventory. Execution counts include repeated human baselines.
| Measure | Minimum | Median | Mean | 95th percentile | Maximum |
|---|---|---|---|---|---|
| Distance, m | 14.4 | 298.0 | 317.0 | 675.9 | 877.7 |
| Elevator rides | 0 | 1 | 1.21 | 2 | 3 |
| Floors crossed | 0 | 2 | 2.86 | 8 | 13 |
| Human travel, min | 0.45 | 7.49 | 7.78 | 13.52 | 18.07 |
| Robot travel, min | 0.60 | 8.83 | 9.25 | 16.34 | 22.12 |
TABLE IV : Distribution of the 1,913 empirical origin and destination routes.
| Input | Low | High |
| Demand multiplier | 0.75 | 1.25 |
| Robot reliability | 0.950 | 0.995 |
| Coordination time, min | 0.5 | 4.0 |
| Congestion sigma | 0.10 | 0.35 |
| Elevator cycle, min | 1.2 | 6.0 |
| Human speed, m/min | 55 | 90 |
TABLE V : Latin hypercube input ranges.
| Policy | Fleet | Late or incomplete (%) | Urgent late (%) | Idle min | Human support |
|---|---|---|---|---|---|
| Human only | 0 | 6.60 | 3.79 | 2068.3 | 0.0 |
| ARAD | 1 | 6.83 | 4.21 | 2059.7 | 29.4 |
| ARAD | 2 | 6.98 | 4.62 | 2056.8 | 56.4 |
| ARAD | 4 | 7.42 | 5.44 | 2052.7 | 100.1 |
| Deadline feasible | 1 | 6.85 | 4.20 | 2062.0 | 29.5 |
| Deadline feasible | 2 | 6.92 | 4.40 | 2058.4 | 54.3 |
TABLE VI: Revised primary results from 30 paired replications. Idle and support values are minutes per 100 nominal beds per measured day.
Fig. 3 : Primary deadline outcomes. Points are means and bars are 95 percent Monte Carlo confidence intervals across 30 replications. The dotted line is the paired human only mean.
| Class | Priority | Deadline | Load | Receive |
|---|---|---|---|---|
| Specimen | 1 | 25, 45, 60 | 1.0 | 0.8 |
| Medication | 2 | 35 | 1.4 | 1.0 |
| Supply | 4 | 120 | 2.0 | 1.5 |
| Instrument | 2 | 45 | 2.5 | 1.5 |
TABLE VII : Assumed task attributes. Times are minutes.
Fig. 4 : Human only lateness under three assumed pooled elevator capacities. The source count mix and 25 minute specimen deadline are fixed. Bars show pointwise 95% Monte Carlo intervals over 30 replications.
| Roster | Task mix | Specimen deadline (min) | Human only (%) | ARAD difference | Proximity difference |
|---|---|---|---|---|---|
| High | Source | 25 | 55.11 | +0.04 | +0.24 |
| High | Source | 45 | 43.82 | +0.19 | +0.21 |
| High | Source | 60 | 38.26 | +0.31 | +0.23 |
| High | Balanced | 25 | 46.27 | -0.16 | +0.01 |
| High | Balanced | 45 | 41.04 | +0.01 | -0.06 |
| High | Balanced | 60 | 38.44 | -0.10 | -0.09 |
TABLE VIII : Overall late or incomplete requests with two assumed elevator servers in 30 replications. Differences are percentage points relative to the paired human only case.
Fig. 5 : Hybrid policy minus human only late or incomplete fraction, in percentage points. Points are means; bars are pointwise 95% Monte Carlo intervals over 30 paired replications. Task mix, deadline, and roster are assumed scenarios.
| Roster | Task mix | Specimen deadline (min) | Human only (%) | ARAD difference | Proximity difference |
|---|---|---|---|---|---|
| High | Source | 25 | 3.18 | +0.11 | +0.03 |
| High | Source | 45 | 0.22 | +0.03 | -0.00 |
| High | Source | 60 | 0.16 | +0.01 | +0.00 |
| High | Balanced | 25 | 1.61 | +0.05 | +0.00 |
| High | Balanced | 45 | 0.25 | +0.01 | +0.00 |
| High | Balanced | 60 | 0.22 | +0.00 | -0.00 |
TABLE IX : Overall late or incomplete requests with four assumed elevator servers in 30 replications. Differences are percentage points relative to the paired human only case.
Fig. 6 : Hybrid policy minus human only lateness under four assumed elevator servers. Points are means; bars are pointwise 95% Monte Carlo intervals over 30 paired replications.
| Bank | Roster | Policy | Coverage (%) | Overall difference | Urgent difference | Idle difference | Support |
|---|---|---|---|---|---|---|---|
| 2 | High | ARAD | 0.15 | +0.04 [-0.07, +0.15] | +0.03 [-0.09, +0.14] | -8.37 [-32.44, +15.70] | 1.7 |
| 2 | High | Proximity | 0.42 | +0.24 [+0.10, +0.38] | +0.22 [+0.09, +0.36] | -60.35 [-90.62, -30.09] | 4.6 |
| 2 | High | Deadline | 0.19 | +0.06 [-0.09, +0.22] | +0.02 [-0.14, +0.19] | -18.95 [-44.71, +6.80] | 2.1 |
| 2 | Moderate | ARAD | 0.82 | +0.22 [+0.15, +0.29] | +0.23 [+0.16, +0.30] | -37.24 [-45.80, -28.67] | 8.9 |
| 2 | Moderate | Proximity | 1.19 | +0.43 [+0.35, +0.51] | +0.43 [+0.34, +0.52] | -74.20 [-83.91, -64.48] | 12.8 |
| 2 | Moderate | Deadline | 0.84 | +0.22 [+0.16, +0.29] | +0.20 [+0.13, +0.27] | -32.09 [-39.83, -24.35] | 8.8 |
TABLE X : Exploratory benchmark comparisons for the source count mix and 25 minute specimen deadline. Bank denotes assumed elevator servers. Every contrast pairs 30 runs by demand seed against human only dispatch. Late differences are percentage points; idle differences and support are minutes per 100 nominal beds per day. Brackets give pointwise 95% Monte Carlo intervals. Coverage is the mean fraction of requests completed by robots. Support excludes worker travel, which is recorded separately.
Appendix figures & tables11 assets
Supplementary material from the paper’s appendix.
Appendix
Fig. 7 : Urgent service failure and transporter utilization across candidate staffing schedules. Each point represents the mean of 30 replications.
| Policy, robots | Late (%) | Late change (pp) | Gross idle change (min) |
|---|---|---|---|
| Adaptive, 1 | 10.33 | -3.40 [-3.51, -3.30] | 51.1 [50.2, 52.1] |
| Adaptive, 2 | 8.50 | -5.24 [-5.38, -5.10] | 79.0 [77.7, 80.4] |
| Adaptive, 4 | 6.48 | -7.26 [-7.44, -7.08] | 105.9 [104.1, 107.7] |
| Overflow, 1 | 10.48 | -3.26 [-3.37, -3.14] | 95.4 [94.5, 96.4] |
| Overflow, 2 | 8.84 | -4.89 [-5.04, -4.74] | 155.3 [153.9, 156.8] |
| Overflow, 4 | 6.71 | -7.03 [-7.21, -6.86] | 232.9 [230.8, 235.0] |
TABLE XI: Superseded comparison before support scheduling and staffing recalibration. Absolute lateness is a percentage; changes are percentage points or gross idle minutes per 100 nominal beds per day. Brackets show pointwise 95 percent Monte Carlo confidence intervals from 100 paired replications. Gross idle intervals exclude robot support work.
Fig. 8 : Superseded paired changes before support scheduling and staffing recalibration. Error bars show pointwise 95 percent Monte Carlo confidence intervals. Contiguous minutes exclude robot support work.
Fig. 9 : Mean late or incomplete fraction by dispatch policy and fleet size at reference demand and robot reliability of 0.98. Error bars show 95 percent confidence intervals.
Fig. 10 : Paired increase in human transporter idle time occurring in intervals of at least 10 consecutive minutes. Values are normalized per 100 beds per day. Error bars show 95 percent confidence intervals.
Fig. 11 : Standardized rank regression coefficients from the global sensitivity analysis. Negative timeliness coefficients indicate inputs that increase the reduction in lateness produced by two overflow robots.
Fig. 12 : Paired effect of two overflow robots under explicit shared elevator contention. Error bars show 95 percent confidence intervals.
Fig. 13 : Paired timeliness and gross contiguous idle time effects under alternative battery and charging infrastructure configurations.
Fig. 14 : Effects of route compatible batching on timeliness, gross contiguous idle time, and robot energy consumption. Labels show batching window and maximum batch size.
Fig. 15 : Effect of mandatory robot decontamination after specimen and instrument transports.
Fig. 16 : Paired effect of overflow robots across alternative virtual hospital layouts with elevator, battery, charging, and decontamination constraints.