Gaussian Mixture Midway-Merge for Object SLAM with Pose Ambiguity

Authors : Jae Hyung Jung, and Chan Gook Park

1. Overview

This is a MATLAB script that implements the Gaussian-mixture invariant EKF with midway-merge in the Chapter 6 of the paper. This demonstration reproduces results of the 0022 sequence in the YCB-Video dataset. Pose detection estimates from CosyPose with a single view are prepared in the cosypose/0022/ folder with the name format as obj_id_imageindex.npy. Note that we use pretrained weights from the authors to estimate camera-object relative poses.

2. Run (YCB-Video example)

Download the YCB-Video toolbox: https://github.com/yuxng/YCB_Video_toolbox
Download the YCB-Video dataset 0022 sequence for a quick demonstration.
Install npy-matlab to read npy files from the pose detector: https://github.com/kwikteam/npy-matlab
Set your path in the main.m and run.

path_toolbox = 'path_to_ycb_video_toolbox';
path_dataset = 'path_to_dataset';

To visualize the estimated masks with the current image, please set isPlot = 1;. But, note that this will slow the process due to the drawing.

3. Citation

If you feel this work helpful to your academic research, we kindly ask you to cite our paper :

@article{Midway_RAL,
  title={Gaussian Mixture Midway-Merge for Object SLAM with Pose Ambiguity},
  author={Jung, Jae Hyung and Park, Chan Gook},
  journal={IEEE Robotics and Automation Letters},
  year={2022},
  publisher={IEEE}
}

4. Acknowledgements

This research was supported by the National Research Foundation of Korea funded by the Ministry of Science and ICT, the Republic of Korea. (NRF-2022R1A2C2012166)

5. License