SCONE

• This GitHub repo is the implementation of "SCONE: A Food Scooping Robot Learning Framework with Active Perception" in CoRL 2023. [openreview] [website]

System Requirements

• Linux (Teseted on Ubuntu 18.04)
• Python 3 (Tested on Python 3.7)
• Torch (Tested on Torch 1.9.1)
• Cuda (Tested on Cuda 11.4)
• GPU (Tested on Nvidia RTX3090)
• CPU (Tested on Intel COre i7-10700)

Setup

• Clone This Repo

$ git clone https://github.com/HCIS-Lab/SCONE.git

• Create Conda Environment

$ cd SCONE
$ conda create -n scone python=3.7
$ conda activate scone

• Install needed package.

1. Please check the website to install pytorch according to your local device.
2. Run pip install -r requirements.txt to install other package.

$ pip install -r requirements.txt

Usage

Check config file to adjust other parameters.

python3 train.py --dataset <data_root>

TODO

• Upload raw & processed dataset
• Complete the Class FoodDataset in load_data.py

Citation

@inproceedings{tai2023scone,
  title={SCONE: A Food Scooping Robot Learning Framework with Active Perception},
  author={Tai, Yen-Ling and Chiu, Yu Chien and Chao, Yu-Wei and Chen, Yi-Ting},
  booktitle={7th Annual Conference on Robot Learning},
  year={2023}
}