Dynamic resource allocation and coordination framework for multi-robot systems in grid-based environments

This paper proposes DyDRAM, a centralized hierarchical coordination framework for lifelong multi-robot operation in grid-based Rack-to-Person warehouses. DyDRAM combines congestion-aware global routing with execution-level conflict handling in a single pipeline. The global planner uses weighted A* with directional constraints, tabu mechanisms, and a heat map cost layer to reduce repeated bottleneck usage, while execution is coordinated through sliding-window reservations and conflict-driven responses to short-horizon conflicts. The framework is evaluated against Windowed Hierarchical Cooperative A* (WHCA*), a Large Neighborhood Search MAPF solver (LNS2), and Enhanced Edge Conflict-Based Search (EECBS) under matched task generation, robot dynamics, and fixed method-specific planning policies. In the baseline long-corridor warehouse layout, DyDRAM with heat map shows the highest observed mean throughput across the tested 10–70-robot range, increasing from approximately 271 to 1126 tasks/hour, while maintaining the lowest observed mean task completion time, from approximately 130 to 215 s. At the highest tested density, the method improves throughput by about 35.5% over the best-performing baseline in throughput and reduces mean task completion time by about 12.8–28.6% relative to the tested baselines. Additional evaluations on a reduced-crossover layout and a scaled-down layout show that the method remains effective within the tested long-corridor warehouse family, with the largest gains appearing when rerouting flexibility is limited and shared-corridor congestion persists. These results are bounded to the fixed simulation protocol, and formal paired significance analysis is reported for the key high-density baseline comparisons.