Reinforcement-Learning-Based 2D Flow Control for Logistics Systems

To address large-scale challenges in current logistics systems, we present an logistics system employing a multi-agent framework grounded on an actuator network named Omniveyor. This design ensures real-time responsiveness and scalability in flow operations. Under the premise of centralized control, we employ Reinforcement Learning (RL) to efficiently control omni-wheel conveyors. Different from traditional path planning, our approach considers the entire platform as an agent, using the package state for observation. Proximal Policy Optimization (PPO) proves to be an effective algorithm for platform-wide control planning. Experimental results demonstrate the system’s capability to accurately deliver packages. Therefore, the RL algorithm allows the omni-wheel platform to autonomously learn and optimize package paths, circumventing the need for traditional controls or path planning methods.