The competition on Flexible Job Shop Scheduling Problems with Worker Flexibility (FJSSP-W) under optional Uncertainty focuses on optimizing production schedules where workers can operate multiple machines but show different skill levels (i.e., processing times) across the tasks. This problem is highly relevant for manufacturing, logistics, and energy operations, where assigning flexible workers affects cost and lead time and where variability in processing rates increases scheduling complexity. The competition further addresses the impact of uncertainty in processing times. This reflects the variability of human workers' processing times to evaluate the robustness of algorithms under more realistic operational conditions.

Although there are many different methods for solving FJSSP-W (e.g. Mathematical Programming, Constraint Programming, etc.), this competition is especially designed for meta-heuristics, such as Evolutionary Algorithms, Swarm Algorithms, or other nature- and physics-inspired methods. Such algorithms often show great results, especially for very complex problem instances. Since there are only a few comparable benchmark instances, such algorithms should demonstrate their capabilities in this competition and compare their performance. In this way, the research community can gain valuable insights into the performance and working principles of contemporary solvers for this class of problems.

The competition provides 30 FJSSP-W benchmark instances with diverse problem characteristics, which rely on well-known FJSSP problem instances. An evaluation method for the FJSSP-W instances as well as a simulation environment for the uncertain scenarios is also specified and provided as Python code. The participants must address at least one of the two different scenarios of the competition, i.e. they may decide to choose the option with or without uncertainty influences on the processing times. The distinction broadens the scope of the competition by allowing existing solvers that were not directly developed for optimization under uncertainty to participate in the competition.

Goals of the competition:

For the first scenario of the competition, the main objective to minimize the makespan across the proposed FJSSP-W test problems. The makespan represents the total time required to complete all tasks in a schedule, i.e. the duration from the start of the first task to the completion of the last task. It represents a critical metric in scheduling because it directly impacts the overall efficiency and performance of a production process. For this reason, it is often the main criterion for a company.

For the second scenario of the competition, the goal is to produce robust schedules towards stochastic processing times. The overall objective remains with the minimization of the makespan. However, the makespan should deteriorate as little as possible when tested subject to the uncertainty simulations. The provided simulation uses different probability distributions for each worker to emulate the fluctuating work performance expected from human workers. The respective parameters are provided by the testing environment and also need to be used for the evaluation of the final result.

As companies are also interested in a balanced distribution of tasks across the workers, we will consider worker balance as a potential tie breaker in case two solvers return equal makespan values on one problem instance.

By doing so, the competition provides a tool for evaluating and comparing scheduling algorithms tailored to FJSSP-W problems. It offers an opportunity to gain insights into the algorithms’ practical relevance in dynamic production environments, where balancing employee flexibility, machine utilization, and production time is essential.

Rules:  

As the competition intends to focus on meta-heuristical solvers of the FJSSP-WF, each solver needs to be executed on each problem instance at least 10 times with a fixed function evaluation budget of 5,000,000. This way, the stochasticity of the solvers is accounted for. After exceeding the time or function evaluation limit, the best candidate solution found needs to be reported for all 10 individual algorithm runs. The feasibility of these final candidate solutions will be evaluated by the organization team after submission of the results. 

Notice that each evaluation of a solution's makespan is counted as a function evaluation. For scenario 2, this means that if multiple simulations are used to average the uncertainty in the objective function, multiple function evaluations must be counted.

For scenario 2, the uncertainty parameters retrieved from the function provided by the environment cannot be changed.

In addition, a short paper or technical report (up to 2 pages following the GECCO formatting requirements and deadlines) outlining the used approached to retrieve the submitted results.