The setting is a robotic mobile fulfillment system in which multiple automated guided vehicles transport pods between storage locations and workstations. The vehicles must be assigned tasks and routed through a parallel-aisle warehouse while avoiding head-on, crossing, and stay-on conflicts, as well as possible deadlocks in narrow aisles. Much prior MAPF work emphasizes collision reduction and travel time, whereas this study adds energy consumption associated with the vehicles’ loading status. The evidence describes a hypothetical RMFS rather than an implementation validated in a physical warehouse.
The first contribution is a collision-avoidance strategy that uses travel time and energy consumption when establishing priorities between conflicting AGVs. The second is a multi-objective task-assignment framework implemented with an NSGA-II approach and a modified Adaptive Large Neighborhood Search procedure, with objectives that include time, energy, collision risk, and travel distance. The work also introduces an ARC operator intended to balance the number of tasks assigned to different AGVs while searching the solution space. The authors state that prior work has not studied energy use in resolving AGV collisions, although the supplied excerpt does not allow the completeness of that literature claim to be independently verified.
The problem is formulated with sets and variables for AGVs, locations, workstations, task locations, loading status, time, and movements on a space-time network. Its four objectives are total running time, maximum running time, total energy consumption, and maximum energy consumption across the AGVs. Routing uses a modified A* search that evaluates location-time states rather than only spatial distance, and collision handling updates the path time when an AGV must wait, step back, or take an alternative route. Task assignment is searched through NSGA-II and ALNS, with multi-objective dominance, neighborhood generation, operator-weight adaptation, and solution acceptance incorporated into the heuristic process. The collision-priority procedure attempts to resolve conflicts and identify deadlocks when vehicles meet inside an aisle.
The routing and priority-rule evaluation compares the proposed rule with a benchmark over thirteen scenarios and reports average values. The task-assignment evaluation includes scenarios with five, eight, and ten AGVs and compares proposed and benchmark versions of ALNS and NSGA-II. The principal indicators are hypervolume, which measures the objective-space region covered by an approximated Pareto front, and covered size of space, which measures dominance relationships between two solution sets. The excerpt provides selected indicator values, but it does not establish all input parameters, random-seed settings, iteration counts, computation times, statistical tests, or physical-robot validation.
For the priority-rule experiments, the displayed hypervolume values are slightly higher for the proposed rule than for the benchmark in each listed condition, suggesting that incorporating energy and time into conflict resolution can improve the reported Pareto approximation. For task assignment, the covered-size results and the paper’s conclusion indicate that the improved ALNS and NSGA-II versions dominate their corresponding existing versions in the reported scenarios. The conclusion further states that NSGA-II performs better than ALNS, attributing part of the difference to the ARC operator’s effort to balance task counts among AGVs. However, the practical significance of the indicator differences, their effect on throughput or charging frequency, and their robustness under heavier or dynamic traffic are not established by the supplied evidence. The authors identify real-time arrivals, battery constraints, and charging policies as important directions for future work.
The paper motivates the problem through the growth of e-commerce and the importance of timely order fulfillment. In a robotic mobile fulfillment system, AGVs move storage pods to workstations, allowing the system to adjust paths dynamically instead of requiring workers to travel throughout the warehouse. Operational decisions include task allocation, path planning, and collision-priority rules, in addition to higher-level decisions such as storage-area sizing and workstation placement. This study focuses on the operational level.
Several AGVs can compete for the same aisle or intersection, creating head-on, crossing, or stay-on conflicts. A conflict-resolution rule that minimizes only delay may overlook the energy consequences of moving a loaded or unloaded vehicle. The paper therefore models routing and task allocation with both aggregate and worst-case travel-time and energy objectives. It seeks feasible, non-conflicting routes and task assignments while preserving a useful set of trade-off solutions rather than a single scalar optimum.
The mathematical formulation represents task assignment, task order, pod pickup and return, loading status, and vehicle movement through a time-expanded network. The four objective functions are total travel time, maximum vehicle travel time, total energy consumption, and maximum vehicle energy consumption. The modified A* procedure evaluates candidate position-time combinations and carries collision-related lost time into subsequent states. Waiting, moving away or stepping back, and rerouting are treated as possible responses to conflicts; a priority rule is used when aisle interactions can otherwise create a deadlock. The task-assignment heuristics use Pareto dominance instead of reducing all objectives to one stated scalar score.
The priority-rule experiment uses thirteen scenarios with different numbers of tasks and AGVs and compares average Pareto-front indicators for the benchmark and proposed routing-priority combinations. The task-assignment experiment evaluates ALNS and NSGA-II in proposed and benchmark forms for cases involving five, eight, and ten AGVs. Hypervolume and covered size of space are used to compare the resulting Pareto sets. Exact warehouse dimensions, processing times, energy coefficients, stopping criteria, hardware, and repetition protocol cannot be recovered from the supplied excerpt because relevant portions are missing.
The priority-rule values shown in the excerpt report higher hypervolume for the proposed rule than for the benchmark in all listed combinations of task and AGV counts. The covered-size material reports strong dominance relationships for the improved task-assignment methods, and the paper summarizes the proposed versions as overwhelmingly dominating the existing versions in the tested cases. The conclusion identifies NSGA-II as better than ALNS and links this result partly to the ARC operator’s task-count balancing. The excerpt does not establish absolute improvements in seconds, energy units, throughput, cost, or statistically significant differences, and the incomplete CSS extract does not support reconstructing every comparison precisely.
The paper extends RMFS path planning and MAPF-style operational planning by treating energy consumption as a first-class objective alongside travel time. Its modified A* routing, energy-aware collision-priority rule, and multi-objective NSGA-II and ALNS task-assignment methods are reported to outperform the benchmarks on computational Pareto indicators for the hypothetical scenarios. The authors especially emphasize the proposed methods’ task-assignment performance and the stronger results obtained with NSGA-II. The evidence supports a promising computational approach, but not a general claim of physical-system superiority.
The supplied text omits portions of the numerical setup and results, so it does not provide enough information to reproduce the experiments fully. The evaluation is computational and hypothetical; it does not demonstrate behavior with real AGVs, communication delays, actuator limits, dynamic order arrivals, finite batteries, charging queues, or uncertain travel times. The energy model’s calibration and the sensitivity of the conclusions to heuristic parameters are also not established in the excerpt. A priority rule that works in the listed scenarios may behave differently in larger, denser, or highly dynamic warehouses.
The work is important as an attempt to connect MAPF collision handling with the energy implications of AGV movement rather than treating every movement as equivalent. Its multi-objective formulation exposes trade-offs between total performance and the worst-performing AGV, while the task-assignment layer connects workload balance to routing conflicts. This perspective may help warehouse designers consider battery use and operational congestion together, although deployment benefits require further validation.
The paper is recommended for researchers working on MAPF, RMFS, AGV routing, warehouse task assignment, and multi-objective meta-heuristics. It is particularly relevant to readers studying NSGA-II, ALNS, Pareto-front indicators, or energy-aware collision avoidance in structured warehouse layouts. It should be read as a computational-method contribution rather than as evidence of field-level performance guarantees. Readers seeking exact reproducibility should consult the complete paper because the supplied excerpt omits essential parameters and some result context.