fig4

Figure 4. Overall architecture of the ATC-N model^[67]. (A) and its components (B). The model features a fused graph made up of three GAT structures: task graph, spatial graph, and semantic graph. This design maximizes the extraction of service scene features and includes a multi-class decoder for predicting task types, corresponding items, and the scene, enabling ATC. ATC-N: Active task cognition network; GAT: graph attention network.