Safety-Aware Adversarial Inverse Reinforcement Learning (S-AIRL) for Highway Autonomous Driving

Safe adversarial inverse reinforcement learning in the paper "Safety-Aware Adversarial Inverse Reinforcement Learning (S-AIRL) for Highway Autonomous Driving"

This repsoitory corresponds to the paper: " Li, Fangjian, John Wagner, and Yue Wang. "Safety-Aware Adversarial Inverse Reinforcement Learning for Highway Autonomous Driving." Journal of Autonomous Vehicles and Systems 1.4 (2021): 041004." -- The hyperparameters have been further tuned to reduce computation load.

Packages:

• python==3.6
• tensorflow==1.14
• numpy==1.16
• gym=0.15.4
• ray=1.2.0
• highway-env==1.2.0

How to run the codes

• come to the root folder of the reporistory
• train benchamrk AIRL: python run_AIRL_combo_highway.py
• CBF-based sampling: python sample_CBF.py
• safety critic training: python run_safety_critic_Q_training.py
• train the SAIRL: run_SAIRL_highway.py

The clips of trained models