IL

Imitation learning algorithms (with PPO [1]):

• ABC [2]
• AIRL [3]
• BC [4]
• GAIL [5]
• GMMIL [6]
• InfoGAIL [7]
• PWIL [8]
• RED [9]

python main.py --imitation [AIRL|BC|GAIL|GMMIL|RED]

Results

PPO

Train	Test

AIRL(s, a)

Train	Test

AIRL(s)

Train	Test

BC

Train	Test

GAIL(s, a)

Train	Test

GAIL(s)

Train	Test

GMMIL(s, a)

Train	Test

GMMIL(s)

Train	Test

RED(s, a)

Train	Test

RED(s)

Train	Test

Acknowledgements

• @ikostrikov for https://github.com/ikostrikov/pytorch-a2c-ppo-acktr-gail

References

[1] Proximal Policy Optimization Algorithms
[2] Adversarial Behavioral Cloning
[3] Learning Robust Rewards with Adversarial Inverse Reinforcement Learning
[4] Efficient Training of Artificial Neural Networks for Autonomous Navigation
[5] Generative Adversarial Imitation Learning
[6] Imitation Learning via Kernel Mean Embedding
[7] InfoGAIL: Interpretable Imitation Learning from Visual Demonstrations
[8] Primal Wasserstein Imitation Learning
[9] Random Expert Distillation: Imitation Learning via Expert Policy Support Estimation