Three-Dimensional Task Allocation of Multiple-Agent Based on Graph Attention Pointer Network

The task allocation problem is a key problem in the study of multi-agent collaboration. The task allocation problem aims to assign tasks to appropriate agents under sequential and logic constraints, so that the quality and efficiency of task completion can be maximized. Several recent studies have shown that attention-based sequence generation models are promising for the task allocation. However, their results are restricted in a two-dimensional space. In this paper, a model based on Graph Attention Pointer Network is proposed for the task allocation problem in a three-dimensional space. The model fully extracts the task features in the three-dimensional space by the graph attention network, then combines the sequence generation model to achieve the task allocation. The attention mechanism in a graph attention network ensures the task allocation performance and the sequence generation method greatly improves the efficiency. Numerical simulations show that the proposed model is suitable for large-scale and dynamic task allocation problems in different three-dimensional scenarios.