[RL] Deep Deterministic Policy Gradient (A.K.A DDPG)

https://github.com/seohyunjun/RL_DDPG 

GitHub - seohyunjun/RL_DDPG: CONTINUOUS CONTROL WITH DEEP REINFORCEMENT LEARNING (a.k.a DDPG)

 

DDPG 

 

*  Continuous Action Space RL 문제 해결 (기존의 DQN discrete action space)

*  DQN에서 actor-critic 사용

*  off-policy 

*  Target Network(=AC) 사용

*  Soft Update(Target Network를 업데이트할 때, parameter update t(tau) 비율 조정)

 

[Example Mujoco_Humanoid-v4] Episode 30 

 

Humanoid-V4 TD3 total reward : 2501.346