Available at: https://digitalcommons.calpoly.edu/theses/3387
Date of Award
6-2026
Degree Name
MS in Industrial Engineering
Department/Program
Industrial and Manufacturing Engineering
College
College of Engineering
Advisor
Mohamed Awwad
Advisor Department
Industrial and Manufacturing Engineering
Advisor College
College of Engineering
Abstract
Mid-market warehouses increasingly pair zone-confined human pickers with a shared fleet of autonomous mobile robots (AMRs) under progressive zoning (PZ), a collaborative system whose operating cost depends on fleet sizing, batch formation, and dispatch policy. Existing fleet-sizing research maximizes throughput, an objective that systematically oversizes fleets under Robotics-as-a-Service (RaaS) subscription pricing because it ignores synchronization loss, the paid idle time incurred when pickers and robots wait for each other at picking locations. This thesis explores fleet sizing as cost-per-line minimization. A full-factorial discrete-event simulation experiment in FlexSim totaling 27,000 runs spans picker staffing levels, AMR-to-picker ratio, batch size, demand density, and dispatch configuration (AMR routing heuristic, mission formation policy, and order queue depth); a semi-open queueing network (SOQN) provides analytical screening, and an XGBoost surrogate with conformal prediction intervals supplies calibrated cost predictions across the continuous design space without further simulation. The cost-per-line surface is U-shaped in the fleet ratio, yielding an interior optimum that shifts from a cost-minimizing ratio of 1 at low demand to 2.5 at high demand. Batch size interacts with the ratio, shifting the optimum by 0.25 to 0.75 units across the parameter space. Picker idle is on average 2.5 times more costly per shift than AMR idle, so the economic impact of synchronization loss falls mainly on the labor side. All three dispatch-heuristic choices have no statistically significant effect on cost-per-line in three-zone progressive zoning under continuous arrivals, a null result that can be traced to three structural mechanisms. The SOQN reproduces simulation cost-per-line within roughly five percent across the design space, the XGBoost surrogate predicts it within three percent under cross-validation, and a publicly deployed decision-support tool serves the validated model hierarchy with calibrated uncertainty.
Award received:
Facilities Planning and Design Track Best Paper at the Institute of Industrial and Systems Engineering Annual Conference