Memory-Adaptive Vision-and-Language Navigation

Keji He, Ya Jing, Yan Huang, Zhihe Lu, Dong An, Liang Wang

Abstract: Highlights•This study is the first work for history noise problem in VLN.•A novel Memory-Adaptive Model is designed to effectively adapts history noises.•Our method shows superior performances on both RxR and R2R datasets.•Our method exhibits strong adaptability across different settings.