crystal-apple / drl Goto Github PK

==================================

https://github.com/openai/spinningup

$ cd DRL $ pip install -e . $ pip install ~/carla/PythonClient (optional) $ pip install opencv-python

==================================

references:

sac is based on:

Soft Actor-Critic: Off-Policy Maximum Entropy Deep Reinforcement Learning with a Stochastic Actor

https://arxiv.org/abs/1801.01290

sac1 is added based on:

Soft Actor-Critic Algorithms and Applications

https://arxiv.org/abs/1812.05905

==================================

ddgp vs sac1

• gym env 'Pendulum-v0':(Minimum_Episode_Return)

sqn experiments on gym env 'LunarLander-v2':

Try trained model on env 'Breakout-ram-v4':

$ python -m spinup.run test_policy ./saved_models/Breakout-ram-v4 -d -l 20000

More experiments: https://mp.weixin.qq.com/s/-ZWj-uw5wWWhGy3B08Xk3Q

INFOBOT: TRANSFER AND EXPLORATION VIA THE INFORMATION BOTTLENECK

• code: to be released.

Unsupervised Meta-Learning for Reinforcement Learning

• https://arxiv.org/pdf/1806.04640.pdf

DIVERSITY IS ALL YOU NEED: LEARNING SKILLS WITHOUT A REWARD FUNCTION (DIAYN)

• https://arxiv.org/pdf/1802.06070.pdf
• code: https://github.com/ben-eysenbach/sac

Model-Agnostic Meta-Learning for Fast Adaptation of Deep Networks (MAML)

• https://arxiv.org/abs/1703.03400