Contrastive RL: Multi skill Reinforcement Learning

🚀 Installation

git clone https://github.com/kkugosu/RL_META.git

🖥️ Experiment

python executable.py

Experiment Parameters

You can choose from the following parameters:

• environments
• policy
• control
• batch size
• memory reset time
• train time per memory
• learning rate
• eligibility trace
• done penalty
• load previous model (not = 0, yes = 1)

🎯 Algorithms

• DIAYN
• VIC
• VALOR
• VISR
• SMM
• EDL
• DADS
• APS
• CIC

🌍 Experiment Environments

• Cartpole
• Hopper

📦 Requirements

• Gym
• Mujoco
• Python >= 3.8
• Pytorch >= 1.12.0
• Numpy

📄 Papers & References

• VIC: https://arxiv.org/abs/1611.07507
• DIAYN: https://arxiv.org/pdf/1802.06070
• VALOR: https://arxiv.org/abs/1807.10299
• VISR: https://arxiv.org/abs/1906.05030
• SMM: https://arxiv.org/abs/1906.05274
• DADS: https://arxiv.org/abs/1907.01657
• EDL: https://arxiv.org/abs/2002.03647
• APS: https://arxiv.org/abs/2108.13956
• CIC: https://arxiv.org/abs/2202.00161