The intent of this library is to provide a standardised framework and reference implementation for online, dense recognition and quantification of human emotions through multiple sensor modalities, to form a reward function.
This reward function may then be used for one or both of:
- Evaluating and comparing the behviour of social robots
- Performing reinforcement learning in social robotics
For more information, please refer to (insert citation)
To get started, run:
git clone --recurse-submodules [email protected]:TomKingsfordUoA/social-reward-function.git
cd social-reward-function
sudo ./install_system_dependencies.sh
pip install .
srf -h
To change branches and update submodules:
git checkout <branch>
git submodule update --init
Speech emotion detection is trained using the following datasets:
These datasets are collected under relatively sterile laboratory conditions and thus there are understandably some issues generalizing to realistic situations. We have decided to include these models under the expectation that the generation of more realistic datasets will lead to more accurate models in the future.
If you use this work, please cite:
@misc{kingsford2022,
title={A General, Evolution-Inspired Reward Function for Social Robotics},
author={Thomas Kingsford},
year={2022},
eprint={2202.00617},
archivePrefix={arXiv},
primaryClass={cs.RO}
}