Abstract: To solve the problem of including invalid nodes in the constructed language graph due to lacking of some clues in the instruction in vision-and-language navigation (VLN) model, this paper designs clues completion module (CCM)to enrich the information expression ability of invalid nodes. A clues-weighted fusion module (CFM) was designed to differentially fuse three clues types, and the fused clue information was used for action prediction, obtaining a more accurate action score to improve the navigation accuracy. Experimental results on the room-to-room (R2R) dataset show that the proposed method has a significant improvement on the main metrics success rate (SR) and success rate weighted by path length (SPL) of VLN tasks.