Strategy Acquisition of the 3-D Packing Problem in Multiagent Environment. GA-based Hierarchical Tuning.

Abstract

In this paper, a new approach to solving automatic 3-D packing problems is proposed. This problem is a well-known complex combinatorial problem. In particular, the difficulties of this problem are compounded in a multiagent environment in which there are multiple agents that perform the packing of corresponding containers. Here, we attempt to establish the mechanism by which the 3-D packing strategy is automatically tuned and the appropriate packing solution is obtained in a multiagent environment, by applying genetic algorithms which mimic the process of a natural evolution system. It this paper, a 3-D packing strategy is acquired through hierarchical tuning. This is achieved by an agent strategy tuning layer and a supervisor strategy tuning layer. First, each agent acquires the individual packing strategy related to the corresponding container. An agent packing strategy is controlled by two evaluation functions which dominate the selection of the next allocation position and box. To find the near-optimal strategy, the weighted coefficients of the evaluation functions are tuned by applying genetic operators such as conventional reproduction, crossover and mutation. Next a supervisor strategy is implemented as a priority sequence of agents. The supervisor strategy is also tuned by modified genetic operators for order-based strings. Based on the proposed method, a 3-D packing simulator was constructed and numerical experiments were carried out. The results of the experiments show the usefulness of the proposed method.