22nd AIAI 2026, 16 - 19 July 2026, Chania, Crete, Greece

Fast, Role-Aware Heuristics for the DynStack Crane Scheduling Problem: From Policy Refinement to Practical Gains

Mahmud Md Tanzeem Hasan, Logofatu Doina, Ray Jiban Kumar

Abstract:

  This paper presents a lightweight heuristic solver for dynamic crane scheduling under the DynStack \texttt{CraneScheduling} benchmark. The setting involves two non-passing cranes, finite-capacity stacks, and online vehicle arrivals operating under strict real-time control constraints. Prior role-based coordination methods often struggle with responsiveness, feasibility under frequent replanning, and crane interference. Our approach introduces three constant-time safeguards: (i) request-aware buffer reassignment with virtual height tracking to avoid burying demanded blocks, (ii) a feasibility swap that guarantees end-to-end crane reachability, and (iii) a precedence-preserving alternating merge that balances crane utilization while maintaining non-crossing safety. Experiments on the official CS--T--01 benchmark achieved zero delivery errors, processed all 185 feasible I/O blocks, and substantially reduced parking and service times compared to earlier role-based strategies. The solver ranked first in the GECCO 2025 DynStack competition. The results show that carefully designed deterministic heuristics can remain highly effective for real-time crane scheduling under strict online control constraints.  

*** Title, author list and abstract as submitted during Camera-Ready version delivery. Small changes that may have occurred during processing by Springer may not appear in this window.