Semantic Cost-Based Task Allocation for Heterogeneous Multi-Robot Systems: Load-Dependent Route and Payload-Margin Costs

Heterogeneous robot fleets transport items that differ in shape and weight, and robots differ in what they can handle and where they can pass. Conventional task allocation methods often compute allocation costs and routes on a single load-invariant spatial model, which may overlook changes in place traversability after loading. Payload capacity typically enters allocation only as a feasibility or ranking condition, so light tasks may be allocated to high-capacity robots also needed for heavier ones. Taking the feasibility information from semantic reasoning as input, this study combines a route cost on load-state-dependent traversability graphs with a normalized payload-margin cost in a makespan-oriented sequential allocation procedure. In ablation experiments, load-invariant route planning produced non-traversable loaded routes for approximately one quarter of the tasks, whereas load-state-dependent planning and validation produced none, as expected by construction. Against a violation-free variant with load-invariant costs and no payload term, the proposed configuration shortened the makespan by 6.7–20.1%, mainly through the payload-margin term; its route cost made the predicted completion times consistent with the planned routes. The reduction persisted on a larger layout and with congested loading and unloading places. In ROS 2 and Gazebo, eight robots followed the planned place sequences in both completed runs.