Joint vs. Modular Training in Job Shop Scheduling with Transport

A new arXiv preprint (2604.24117) investigates the coordination gap between joint and modular multi-agent reinforcement learning for job-shop scheduling with transportation resources. The study systematically analyzes when joint training—simultaneous training of job and automatic guided vehicle scheduling agents—is necessary for optimal performance, versus modular training where each agent is trained independently and integrated post-hoc. Through sensitivity analysis of resource scarcity and temporal dominance, the authors quantify the conditions under which joint training outperforms modular approaches. The work addresses a gap in prior research, which has focused on novel cooperative architectures without examining when joint training is essential. This research is relevant to decentralized factories and high-performance manufacturing.