Abstract:

A complex visual navigation task puts an agent in different situations which call for a diverse range of visual perception abilities. For example, to "go to the nearest chair'', the agent might need to identify a chair in a living room using semantics, follow along a hallway using vanishing point cues, and avoid obstacles using depth. Therefore, utilizing the appropriate visual perception abilities based on a situational understanding of the visual environment can empower these navigation models in unseen visual environments. We propose to train an agent to fuse a large set of visual representations that correspond to diverse visual perception abilities. To fully utilize each representation, we develop an action-level representation fusion scheme, which predicts an action candidate from each representation and adaptively consolidate these action candidates into the final action. Furthermore, we employ a data-driven inter-task affinity regularization to reduce redundancies and improve generalization. Our approach leads to a significantly improved performance in novel environments over ImageNet-pretrained baseline and other fusion methods.

Our Model:

Fusion-weight patterns:

We observe that the learned fusion weight exhibits interesting patterns. Namely, in narrower spaces like corridors, the weight skews towards 3D geometric features; while in open spaces like living room with different objects, the weight skews towards semantic feature.

Bibtex:

@inproceedings{situational2019,

title={Situational Fusion of Visual Representation for Visual Navigation},

author={William B. Shen and Danfei Xu and Yuke Zhu and Leonidas J. Guibas and Li Fei-Fei and Silvio Savarese},

booktitle={IEEE International Conference on Computer Vision (ICCV)},

year={2019},

}