An Intention Inference Method for BiGRU Integrating Multi-head Self-Attention in Share Control

Human-machine cooperative systems based on human-artificial intelligence (AI) in share control have made substantial progress in recent years. Intention inference, a core component of share control, determines the degree of accuracy with which artificial intelligence agents recognize human intent. A good intent reasoning algorithm can enable the intelligent body to efficiently assist the user in accomplishing the target task. However, existing intent inference methods cannot flexibly capture the sequential relationships among user control trajectories. In this paper, we propose a novel intent inference method named GEMH-BiGRU that incorporates a multi-head attention mechanism and bidirectional gating recurrent unit (BiGRU). The BiGRU outputs hidden states containing contextual information, and the multi-head self-attention mechanism parses the feature connections between target attributes and combines them by means of weights, which fully considers the correlation and sparsity of trajectory data. We create intelligent agents that mimic different human strategies and employ a practical evaluation method. Experimental results show that all our models have strong consistency and prediction results.