Pricing
Fabricatio-Rl: A Reinforcement Learning Simulation Framework For Production Scheduling
This work introduces FabricatioRL, an RL compatible, customizable and extensible benchmarking simulation framework that can interface with both traditional approaches and RL, and ensure that generic production setups can be covered, and experiments are reproducible.
Abstract
Production scheduling is the task of assigning job operations to processing resources such that a target goal is optimized. constraints on job structure and resource capabilities, including stochastic influences, e.g. job arrivals, define individual problems. Reinforcement learning (RL) solvers are adaptive and potentially robust in highly stochastic settings. However, benchmarking RL solutions for stochastic problems is challenging, requiring the simulation of complex production settings while guaranteeing reproducible stochasticity. No such simulation is currently available. To cover this gap, we introduce FabricatioRL, an RL compatible, customizable and extensible benchmarking simulation framework. Our contribution is twofold: We first derive requirements to ensure that generic production setups can be covered, the simulation framework can interface with both traditional approaches and RL, and experiments are reproducible. Then, we detail the FabricatioRL design and implementation satisfying the obtained requirements in terms of framework input, core simulation process, and the interface with different scheduling systems.