daniellawson9999/compositional_reinforcement_learning

Deep reinforcement learning-basedskill transfer and composition method

COMPOSING TASK-AGNOSTIC POLICIES WITH DEEP REINFORCEMENT LEARNING

• Requirements:

1. Rllab
2. Tensorflow
3. mujoco

To train composite model from scratch, run:

1. To simulate "ant-cross-maze", run:

python mujoco_am_sac.py --log_dir="/path-to-crl-code-folder/composition_sac_code/ant-maze" --domain="ant-cross-maze"

2. To simulate "ant-random-goal", run:

python mujoco_am_sac.py --log_dir="/path-to-crl-code-folder/composition_sac_code/ant-rgoal" --domain="ant-random-goal"

3. To simulate "cheetah-hurdle", run:

python mujoco_am_sac.py --log_dir="/path-to-crl-code-folder/composition_sac_code/cheetah-hurdle" --domain="cheetah-hurdle"

4. To simulate "pusher", run:

python mujoco_am_sac.py --log_dir="/path-to-crl-code-folder/composition_sac_code/pusher" --domain="pusher"

References

@inproceedings{
qureshi2020composing,
title={Composing Task-Agnostic Policies with Deep Reinforcement Learning},
author={Ahmed H. Qureshi and Jacob J. Johnson and Yuzhe Qin and Taylor Henderson and Byron Boots and Michael C. Yip},
booktitle={International Conference on Learning Representations},
year={2020},
url={https://openreview.net/forum?id=H1ezFREtwH}
}