A reinforcement learning algorithm for the 2D-rectangular strip packing problem

Abstract

The 2D packing problem is categorized as one branch of the cutting and packing problems, which is widely spread in the manufacturing industries. Over the years many meta­heuristics have been proposed and applied on the packing problem. Recently, the approach combined with machine learning serves as a novel paradigm for solving the combinatorial optimization problem. However, the machine learning approaches have very limited literature reports on the appliance of the packing problem. We propose a reinforcement learning method for the 2D-rectangular strip packing problem. The solution is represented by the sequence of the items and the layout is constructed piece by piece. We use the lowest centroid placement rule for the piece placement, then a Q-learning based sequence optimization is applied. Three groups of conditions are set for the testing, the computational results show the Q-learning approach has good effect on the compaction of the layout.