This is an implementation of Deep Deterministic Policy Gradient (DDPG) in bullet Gym using PyTorch.

• Python 3.6.2
• pytorch 0.2.0
• gym
• tensorboardX-1.0
• pybullet (if you want to train agents for bullet env)

• here is a simple example to train CartPole with high efficiency:

$ cd base

$ python main.py --debug --discrete --env=CartPole-v0 --vis

• you can use this to understand usage of each argument:

$ python main.py --help

• some explanation of important arguments:

--debug: print the reward and some other information

--discrete: if the actions are discrete rather than continuous

--vis: render each action (but it would slow down your training speed)

--cuda: train this task using GPU

--test: testing mode

--resume : load model from the path

# run HalfCheetah-v1 with default parameters
cd ./baselines
python main.py 

• MemphiSqrt
• hzwer
• KuribohG