Scheduling Method of Robotic Flexible Flow Shops based on Petri Nets and Masked-PPO

Abstract

With the rapid development of the smart industry, robotic flexible flow-shop (RFFS) has been widely adopted in intelligent manufacturing systems due to its strong capability to support customized, multi-variety, and small-batch production modes. To enhance both solution quality and computational efficiency in addressing the scheduling problem of RFFS, this paper proposes a scheduling method that combines Petri nets with a Masked Proximal Policy Optimization (Masked-PPO) algorithm. The production process is modeled by timed Petri nets, and the scheduling problem is formulated as a Markov decision process. The state space, action space, and reward function are set accordingly. Due to complex process flows and resource constraints, many actions in the complete action space may be infeasible in a given state, leading to inefficient exploration. To tackle this issue, we integrate an action masking mechanism into the PPO algorithm. By incorporating a deadlock-avoidance policy based on Petri nets, infeasible actions are dynamically filtered, thereby restricting policy selection to feasible actions. Experimental results demonstrate that the proposed MaskedPPO algorithm reduces invalid decisions, improves training stability, shortens episode lengths, and achieves competitive performance across various problem scales.