Interacting with multi-agent systems through intention field based shared control methods

In this paper, a two-layer shared control system is proposed to facilitate interactions between human and multi-agent systems (MAS). The lower layer is an autonomous system that performs a formation task, and the upper layer is an intention field network that generates control intentions from human inputs for the lower layer. These two layers are connected by a priority-based policy blending algorithm. The use of an intention field network is the key ingredient of the proposed method, as it enables fast propagation of human intentions in the multi-agent system, making the overall system more flexible. Theoretic proofs are given that guarantee the stability of the formation task. A comparison experiment is carried out that shows the advantages of using intention field networks in a multi-agent shared-control system.