data_parser.md

Downloading Datasets

To download SCAND please follow this link and download the bag files. And download MuSoHu bag files from here.

Installing Dependencies

The parser requires Python>=3.9 for type annotations. pyntcloud package works best with pandas==2.0.1.

Parsing the data

It is recommended to place all the bag files inside the data directory as depicted in the project structure below. Otherwise, you need to change bags_dir variable in the parser config file. You can also change other parameters inside the config files.

Project structure:

 .
├──  docs
│   └──  data_parser.md
├──  LICENSE.txt
├──  README.md
├──  requirements.txt
├──  run.sh
└──  social_nav
    ├──  __init__.py
    ├──  conf
    │   ├──  config.yaml
    │   ├──  musohu_parser.yaml
    │   └──  scand_parser.yaml
    ├──  data
    │   ├──  musohu
    │   │   └──  03112023_mn_dc_night_1_casual.bag
    │   ├──  processed
    │   │   ├──  A_Jackal_AHG_Library_Thu_Nov_4_16_0
    │   │   │   ├──  point_cloud
    │   │   │   ├──  rgb
    │   │   │   └──  traj_data.pkl
    │   │   ├──  mn_dc_night_1_casual_0
    │   │   │   ├──  depth
    │   │   │   ├──  point_cloud
    │   │   │   ├──  rgb
    │   │   │   └──  traj_data.pkl
    │   └──  scand
    │       └──  A_Jackal_AHG_Library_Thu_Nov_4_16.bag
    ├──  main.py
    ├──  misc_files
    │   └──  FUTURAM.ttf
    ├──  model
    │   ├──  __init__.py
    │   ├──  data_loader.py
    │   └──  net.py
    └──  utils
        ├──  __init__.py
        ├──  helpers.py
        ├──  io.py
        ├──  musohu_parser.py
        ├──  nn.py
        ├──  parser.py
        ├──  parser_utils.py
        ├──  scand_parser.py
        └──  vision.py

To run the MuSoHu parser, from the root directory of the project run:

./run.sh parser musohu

To run SCAND parser, change the parser argument to scand like:

./run.sh parser scand

We only store the front-facing camera for the Spot in SCAND, so both MuSoHu and SCAND have the same interface. The only difference is that SCAND does not contain depth data.

Creating Samples

To create samples from parsed bags, run the following command:

./run.sh parser sampler

The sampler uses the parser config file to create samples. The sampler uses the observation length (obs_len) and prediction length (pred_len) from the parser config file to create samples. It uses the directory for storing the pickle file from parsed_dir.