MiC

Code of the ICCV 2023 paper :
March in Chat: Interactive Prompting for Remote Embodied Referring Expression

Abstract

Many Vision-and-Language Navigation (VLN) tasks have been proposed in recent years, from room-based to object-based and indoor to outdoor. The REVERIE (Remote Embodied Referring Expression) is interesting since it only provides high-level instructions to the agent, which are closer to human commands in practice. Nevertheless, this poses more challenges than other VLN tasks since it requires agents to infer a navigation plan only based on a short instruction. Large Language Models (LLMs) show great potential in robot action planning by providing proper prompts. Still, this strategy has not been explored under the REVERIE settings. There are several new challenges. For example, the LLM should be environment-aware so that the navigation plan can be adjusted based on the current visual observation. Moreover, the LLM planned actions should be adaptable to the much larger and more complex REVERIE environment. This paper proposes a March-in-Chat (MiC) model that can talk to the LLM on the fly and plan dynamically basexzd on a newly proposed Room-and-Object Aware Scene Perceiver (ROASP). Our MiC model outperforms the previous state-of-the-art by large margins by SPL and RGSPL metrics on the REVERIE benchmark.

TODOs

Demonstration Set
ROASP Code
LLM Code

Prerequisites

Installation and Data Preparation

Please follow the installation instructions in DUET to set up the environment and download the dataset.

Trained Weights

Download the trained weights from here.

Room-and-Object Aware Scene Perceiver (ROASP)

cd ROASP

To predict the room type

python process-clip-room-ViTB32.py

To predict the object type

python process-clip-obj-ViTB32.py

Citation

Please cite our paper:

@InProceedings{Qiao_2023_MiC,
    author    = {Qiao, Yanyuan and Qi, Yuankai and Yu, Zheng and Liu, Jing and Wu, Qi},
    title     = {March in Chat: Interactive Prompting for Remote Embodied Referring Expression},
    booktitle = {Proceedings of the IEEE/CVF International Conference on Computer Vision (ICCV)},
    month     = {October},
    year      = {2023},
    pages     = {15758-15767}
}

Acknowledgement

We thank the developers of DUET, language-planner for their public code release.

YanyuanQiao/MiC

MiC