A Deep Reinforcement Learning Method for 2D Irregular Packing with Dense Reward

In this work, we present a Deep Reinforcement Learning (Deep RL) method to achieve 2D irregular packing, which has a wide range of applications in the logistics industry. We trained a Deep RL agent to predict the optimal placement location in the container while maximizing the available area. Compared with previous algorithms using Deep RL, our proposed definition of the reward function was applied at each step (dense reward), providing the agent with more immediate feedback and enabling faster learning. To the best of the authors' knowledge, this is the first time that a dense reward has been applied to solve the 2D irregular packing problem. Simulation results show that our method can successfully complete the online 2D irregular packing task, achieving promising performance in terms of volume efficiency and number of packed pieces. We believe that our proposed technique provides new insights into the use of deep learning to solve irregular packing problems, and it has practical significance for improving the productivity of factories and logistics warehouses.