LangNav: Language as a Perceptual Representation for Navigation
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2023
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917626560643072 |
|---|---|
| author | Pan, Bowen Panda, Rameswar Jin, SouYoung Feris, Rogerio Oliva, Aude Isola, Phillip Kim, Yoon |
| author_facet | Pan, Bowen Panda, Rameswar Jin, SouYoung Feris, Rogerio Oliva, Aude Isola, Phillip Kim, Yoon |
| contents | We explore the use of language as a perceptual representation for vision-and-language navigation (VLN), with a focus on low-data settings. Our approach uses off-the-shelf vision systems for image captioning and object detection to convert an agent's egocentric panoramic view at each time step into natural language descriptions. We then finetune a pretrained language model to select an action, based on the current view and the trajectory history, that would best fulfill the navigation instructions. In contrast to the standard setup which adapts a pretrained language model to work directly with continuous visual features from pretrained vision models, our approach instead uses (discrete) language as the perceptual representation. We explore several use cases of our language-based navigation (LangNav) approach on the R2R VLN benchmark: generating synthetic trajectories from a prompted language model (GPT-4) with which to finetune a smaller language model; domain transfer where we transfer a policy learned on one simulated environment (ALFRED) to another (more realistic) environment (R2R); and combining both vision- and language-based representations for VLN. Our approach is found to improve upon baselines that rely on visual features in settings where only a few expert trajectories (10-100) are available, demonstrating the potential of language as a perceptual representation for navigation. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2310_07889 |
| institution | arXiv |
| publishDate | 2023 |
| record_format | arxiv |
| spellingShingle | LangNav: Language as a Perceptual Representation for Navigation Pan, Bowen Panda, Rameswar Jin, SouYoung Feris, Rogerio Oliva, Aude Isola, Phillip Kim, Yoon Computer Vision and Pattern Recognition Artificial Intelligence Computation and Language Robotics We explore the use of language as a perceptual representation for vision-and-language navigation (VLN), with a focus on low-data settings. Our approach uses off-the-shelf vision systems for image captioning and object detection to convert an agent's egocentric panoramic view at each time step into natural language descriptions. We then finetune a pretrained language model to select an action, based on the current view and the trajectory history, that would best fulfill the navigation instructions. In contrast to the standard setup which adapts a pretrained language model to work directly with continuous visual features from pretrained vision models, our approach instead uses (discrete) language as the perceptual representation. We explore several use cases of our language-based navigation (LangNav) approach on the R2R VLN benchmark: generating synthetic trajectories from a prompted language model (GPT-4) with which to finetune a smaller language model; domain transfer where we transfer a policy learned on one simulated environment (ALFRED) to another (more realistic) environment (R2R); and combining both vision- and language-based representations for VLN. Our approach is found to improve upon baselines that rely on visual features in settings where only a few expert trajectories (10-100) are available, demonstrating the potential of language as a perceptual representation for navigation. |
| title | LangNav: Language as a Perceptual Representation for Navigation |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence Computation and Language Robotics |
| url | https://arxiv.org/abs/2310.07889 |