ADVANCING MATERIAL HANDLING SYSTEMS WITH AUTONOMOUS MOBILE ROBOTS THROUGH MACHINE LEARNING AND INTEGRATED CENTRALIZED-DECENTRALIZED CONTROL

As customer demand for personalized items rises, the need for flexible material handling systems (MHSs) becomes increasingly critical to accommodate growing product variety and demand fluctuations. Autonomous mobile robots (AMRs) play a key role in addressing these challenges by operating autonomously and adapting to dynamic environments. However, the decentralized decision-making of AMRs can limit their effectiveness in considering a system-wide perspective. Moreover, as the AMR fleet size increases, system performance loss due to traffic congestion becomes inevitable. To address these challenges, this dissertation investigates AMR operations in MHSs in three steps. First, to explore ideal optimal operation plans for AMRs, we assume a centralized controller managing the entire fleet in a static environment, where all information is known in advance. A mixed-integer linear programming (MILP) model, a genetic algorithm (GA), and a collision avoidance scheduling algorithm are developed to generate centralized solutions. Second, to overcome the myopic nature of decentralized decision-making, a machine learning-based platform is proposed. This platform extracts knowledge from solutions obtained in a static environment from the centralized controller’s perspective and uses it as a decentralized protocol for individual AMR operations in a dynamic environment, where future information is unavailable. Experimental results show that the proposed protocol outperforms conventional dispatching rules in most cases. Lastly, to mitigate system deterioration caused by traffic congestion in MHSs, a job-initiated traffic-aware protocol (JTAP) is proposed. This protocol requires collaboration between centralized and decentralized approaches and consists of traffic zone configurations, potential congestion level measurements, and AMR availability control. Experimental results show that JTAP outperforms the baseline scenario, where traffic is not controlled, and demonstrates its effectiveness in reducing AMR fleet size while maintaining system performance and enhancing MHS resilience to traffic congestion.