1 paper
Yi Wu, Junjie An, Xiao Liu +6
SoftNav introduces a method to embed continuous 3D object representations as soft tokens directly into a frozen vision‑language model, enabling efficient embodied navigation with m…