ABSTRACT. The Flexible Job Shop Scheduling Problem (FJSP) is a complex scheduling challenge widely encountered in the manufacturing industry. Traditional heuristic methods often struggle to adapt to large-scale and dynamically changing production environments. This study proposes a hybrid scheduling approach combining dispatching rules and Deep Reinforcement Learning (DRL), specifically using the Deep Q-Network (DQN) algorithm. By integrating priority-based dispatching mechanisms with DQN, the model effectively balances on-time delivery and total tardiness minimization in dynamic FJSP settings. Experimental results demonstrate that this hybrid approach significantly improves scheduling performance and adaptability over conventional methods.

ABSTRACT. Optical Transport Networks (OTNs) serve as the backbone of modern telecommunications by offering high-capacity and long-distance data transmission. In this work, we propose a novel framework that combines Graph Neural Networks (GNNs) with a Double Dueling Deep Q-Network (D3QN) to comprehensively capture both the complex topological features of OTNs and the dynamic routing decisions required for high-performance operation. Our method is evaluated on representative optical transmission network topologies, and the experimental results demonstrate significant performance improvements in resource utilization, latency, and reliability compared to existing baselines. These findings underscore the potential of synergizing Deep Reinforcement Learning (DRL) with GNNs for large-scale OTN optimization.

ABSTRACT. This study proposes a product allocation method which equalizes the workload of machines by using an an-nealing method. The proposed method determines the products to be re-allocated probabilistically in consider-ation of the standard time obtained by dividing the processing time of all products by the number of machines. We also showed that an annealing method that generates a neighborhood solution using a stochastic product allocation method that takes the standard time into account can obtain a value closer to the optimal solution more efficiently than an annealing method that randomly selects products and machines to generate a neigh-borhood solution.

ABSTRACT. In a scheduling game, each strategic job acts as a player that selects one of several parallel machines. Each job aims to minimize its own completion time while accounting for the decisions of other jobs. A (pure) Nash equilibrium is reached when no job can unilaterally switch machines to improve its completion time. The price of anarchy (PoA) is a standard metric for evaluating the efficiency of such equilibria with certain social optimum. Prior research has investigated decentralized learning in multi-agent systems, where participants adapt their strategies independently. In this paper, we define the social cost as the total sum of job completion times and propose a more generalized framework that accommodates diverse cost and utility models in congestion games, which include scheduling games as a special case. Under the bandit feedback model where each participant only observes the cost of their selected resource without access to unchosen alternatives, we develop a decentralized bandit learning algorithm that ensures rapid convergence to the social optimum.

ABSTRACT. Efficient evacuation strategies are crucial in emergency scenarios, particularly in multi-exit environments where pedestrian dynamics play a significant role. This study investigates the impact of different agent activation methods—Sequential, Random, and Simultaneous—on evacuation efficiency, deadlock resolution, collision frequency and other metrics. Using an agent-based pedestrian movement model, we incorporate realistic pedestrian characteristics such as body diameter, speed based on real-world data. The movement model is velocity-based, handling wall avoidance and inter-agent collisions. Simulations with varying agent populations and model steps provide insights into the optimal conditions for each activation method. The results contribute to a better understanding of evacuation efficiency and congestion mitigation strategies in different environment settings.

ABSTRACT. In recent years, the development of optimization methods in multi-agent systems has been remarkable. We regard a production system as a multi-agent system in which each piece of equipment is represented by an agent, and we have proposed a consensus-based distributed scheduling method (CDBS) using the Alternating Direction Method of Multipliers (ADMM), where agents determine schedules while reaching agreements with neighboring agents. However, many cases exist where the scheduling oscillates and fails to converge; thus, the issue of convergence remains. On the other hand, one of the control techniques in multi-agent systems is pinning control. Pinning control is a method that controls the overall dynamics of a multi-agent system by applying control to a limited number of agents. This paper aims to evaluate the effect of pinning control in CDBS through computational experiments. We apply pinning control to CDBS for the scheduling problem of a rolling system. The results of the computational experiments demonstrate that pinning control can lead CDBS to convergence.

ABSTRACT. The use of EVs for middle-mile deliveries (deliveries between distribution centers) is limited due to their short cruising range compared to gasoline-powered vehicles. The large-scale introduction of charging stations is necessary to use EVs for middle-mile deliveries. Therefore, we construct a model for optimizing the placement of charging stations. In the model, we introduce time windows for each charger to account for variations in charging demand. Moreover, we consider both the cost of charger installation and the operating costs of EVs to reveal the trade-off between infrastructure investment and vehicle running costs.

ABSTRACT. In the logistics industry, the shortage of truck drivers has led to a severe decline in transportation capacity. This study proposes an optimization method for delivery planning by introducing autonomous delivery robots into small-parcel delivery operations and coordinating them with trucks. In this approach, clustering methods are used to determine the allocation of delivery locations, and approximate solution methods are applied to optimize delivery routes. Furthermore, computational experiments are conducted to verify the improvement in delivery efficiency.

ABSTRACT. This research proposes a genetic algorithm to optimize the job-shop scheduling problem in n×n scale factories, incorporating time-of-use electricity pricing. The study develops a mathematical model that simultaneously minimizes both make-span and power consumption. Through genetic operations and a specialized decoding process that considers job urgency and electricity costs, the approach implements soft constraints for time-of-use electricity pricing windows. This strategy effectively reduces power consumption while minimizing potential make-span extensions, achieving a balanced optimization between production efficiency and energy costs.

ABSTRACT. This paper considers the problem of minimizing the carry-over effects value in round-robin tournaments, which is to create a fair schedule in sports scheduling. We formulate the problem as a 0-1 convex quadratic optimization problem and solve it using a QUBO solver. Our approach improved the best known result of the carry-over effects value when the number of teams in a tournament is 14.

ABSTRACT. The total early work criterion, as complementary to total late work, reflects the total benefit resulted from scheduling jobs relative to their specified due dates. This study investigates the problem of maximizing total early work in distributed flow shops, which are composed of multiple identical parallel two-stage flow shops. To tackle this challenge, we first present a mixed-integer programming (MIP) model that formalizes and addresses the problem. While the general problem is computationally intractable, we explore the complexity of a simplified scenario involving a single flow shop and establish that it remains strongly NP-hard. For cases with a fixed number of due dates, we introduce a pseudo-polynomial time dynamic programming algorithm. Currently, we are designing heuristic approaches, including a genetic algorithm and an ejection-chain algorithm, to generate approximate solutions. Computational experiments will be conducted to examine the performances solution quality and rune time, of the proposed solution approaches.

ABSTRACT. When it is not possible to run a full round-robin tournament due to the availability of time slots or places, a tournament schedule is constructed in parts of a round-robin tournament. In this schedule, it is assumed that matches assigned to the same time slot are played simultaneously. Thus, each team plays at most one match in each time slot. Considering the fairness of the schedule, this paper addresses the problem of minimizing the difference in the number of matches for each team and of minimizing the number of matches with rest mismatch.

ABSTRACT. In this paper, we treat a last-mile parcel delivery scheduling model of minimizing the makespan for a truck-drone system, which consists of multiple drones and a single carrier truck. We are given a fixed truck route,in which we are also given a finite set of stops. The carrier truck visits the stops in a prescribed order of them, and at a stop, it can launch~(resp., retrieve) a drone to~(resp., from) a customer. Each customer must be served, i.e., must be delivered a parcel, exactly once by a drone launched from the carrier truck at a stop. For a sortie of a drone with a takeoff stop, the following restrictive options of selecting a landing stop have been considered: (i) the landing stop must be identical to the takeoff stop, (ii) the landing stop must be either identical to the takeoff stop or the next stop of the takeoff one, and (iii) the landing stop can be any non-earlier stop of the takeoff stop in a given truck route. A single restrictive option of selecting a landing stop has generally been imposed on all given drones, while in our model, the drones have a mixture of two sortie options. More precisely, we are given a specified drone, namely libero, with the third option and the other drones have the second option. In this paper, we obtain a mixed integer program~(MIP) of the last-mile parcel delivery scheduling model of the truck-drone system with a libero role by modifying an existing MIP. We also conduct numerical experiments to demonstrate the solution of the %presented modified MIP by utilizing a well-known solver, and report the results.

ABSTRACT. In modern intelligent warehousing and logistics systems, the multi-AGV conflict-free path planning problem significantly impacts operational efficiency. This study proposes a conflict-free path planning and scheduling approach based on Dijkstra’s algorithm and dynamic task allocation to address the task-to-AGV ratio imbalance. The method determines the shortest paths while a time window mechanism prevents conflicts through dynamic adjustments. The results demonstrate the approach reduces total path length and task completion time, ensuring smoother AGV navigation. It offers significantly improved efficiency, flexibility, and scalability, making it a viable solution for automated logistics and intelligent warehousing.

ABSTRACT. We treat an issue of routing and scheduling for a last-mile parcel delivery system composed of a single carrier truck and multiple identical drones. In this paper, we first propose a mixed integer program of this issue, which is an extension of an existing integer program for a fixed truck route model of such a truck-drone system. The objective is to minimize the sum of the total duration over available truck stops and a weighted travel time of the carrier truck, i.e., it is to minimize a generalized makespan. We also demonstrate the solution quality of the proposed mixed integer program empirically by utilizing a mixed integer programming solver, and report the results.

ABSTRACT. This paper considers order batching and batch assignment sequencing problems to solve unbalanced allocation and unscheduled picking. To solve the problems, we develop an optimization model for order-picking synchronization with the objective of minimizing the makespan time of picking orders to improve the lead time, stagger accuracy, and productivity. This study develops a multilevel linear regression model to predict the picking batch execution time and later using it in the optimization model. As a result, the optimization model has successfully reduced the makespan time and improved the lead time, picking stagger accuracy, and productivity by subsequently 7.21%, 67.4%, and 62.9%.

ABSTRACT. This study addresses a coil stacking problem arising in the steel industry. Steel coils are stacked in two or three layers in a warehouse. Unlike containers that can be stacked vertically, a steel coil must be supported by two coils beneath it because of its cylindrical shape. This study focuses on a class of coil stacking problems, aiming to load incoming steel coils into a storage area so that no coil blocks the retrieval of other coils. This problem is formulated as an integer programming problem. Next, simple heuristics applicable to both container and coil stacking problems are developed, which simultaneously prove polynomial solvability of special classes of the problems. A computational study is conducted to investigate the similarities and differences between the coil and container stacking problems.

ABSTRACT. Determining the number of robots to be introduced is a crucial issue. A few robots reduce work efficiency, whereas several robots reduce their utilization rate. In this study, a 0-1 integer programming problem was formulated to determine the work sequence that minimizes the picker’s waiting time. The number of robots satisfying the work efficiency and utilization rate requirements was determined by calculating the work efficiency and utilization rate for each number of robots.