Code for Learning World Transition Model for Socially Aware Robot Navigation accepted in ICRA2021.

Paper is available here.

Video is available here.

Presenting video is available here on Bilibili.

git clone https://github.com/YuxiangCui/model-based-social-navigation.git
catkin_make -j6
source devel/setup.bash

roslaunch model_free_version start.launch (FOUR-AGENT)

for policy performance testing or model free training

• main.py (4 agent main function)
• policy.py (policy network)
• environment_four.py (4 agent environment)
• agent.py (agent's states, reward, action...)
• utils.py (replay buffer)

roslaunch model_based_version start.launch

• main_mbpo.py (1/4 agent main function)

• env_sample.py

• environment_one_agent.py (real 1 agent environment)

• env_sample_four.py

• environment_four_agent.py (real 4 agent environment)

• agent.py (agent's states, reward, action...)

• replay_buffer_env.py (real data replay buffer)

• replay_buffer_model.py (virtual data replay buffer)

• policy.py (policy network)

• transition_model.py (world transition model)

• ensemble_model_train_mcnet_all.py

• env_predict.py (virtual environment)

Code references MBPO. TD3.

If you use our source code, please consider citing the following:

@article{cui2020learning,
  title={Learning World Transition Model for Socially Aware Robot Navigation},
  author={Cui, Yuxiang and Zhang, Haodong and Wang, Yue and Xiong, Rong},
  journal={arXiv preprint arXiv:2011.03922},
  year={2020}
}