@inproceedings{cai2021reinforcement,
title={Reinforcement learning based temporal logic control with maximum probabilistic satisfaction},
author={Cai, Mingyu and Xiao, Shaoping and Li, Baoluo and Li, Zhiliang and Kan, Zhen},
booktitle={2021 IEEE International Conference on Robotics and Automation (ICRA)},
pages={806--812},
year={2021},
organization={IEEE}
}
The agent have been trained on the task: go to goal-1 and then to goal-2 with maximum probability
Clone this repository and install the coressponding dependencies:
git clone https:https://github.com/mingyucai/E-LDGBA_RL
Execute the example:
python3 main.py --env 'SlipperyGrid' --layout 'layout_1' --property 'g1-then-g2'
This benchmark shows an E-LDGBA built from LDGBA. As for more samples, the LTL can be converted to LDGBA or LDBA by excellent tool OWL, which is available at (https://owl.model.in.tum.de/try/).