Training OpenAI Gym Half Cheetah environment using Soft-Actor-Critic with maximum entropy algorithm

Contributors:

Tej Kiran

Ji Liu

Required libraries and tested environment settings

Please make sure to create the environment using the provided conda environment file

-Create conda enviornment
    conda env create -f EnvSAC.yml

Instructions to run the code:

Step 1: Clone the repositiory

Step 2: Open terminal, activate the python environment and cd to the above "code" location

Step 3: To train the network run "python3 sac_train.py"

Step 4: To train the network run "python3 sac_test.py"

Training

The Model has been trained in 3 different stages with the temperature parameter alpha reset at the beggining of every stage

• Stage 1 : 500 complete episodes
• Stage 2 : 5000 episodes each terminated at 100th step
• Stage 3 : 1000 episodes each terminated at 500th step

Results

• The normalized reward plot is as shown below

• The achieved results in the simulation is as shown in the below

Simulation.Results.mp4

Errors and solutions

To avoid path errors when training on nvidia GPU, please run the following commands

- export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:~/.mujoco/mujoco210/bin
- export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/lib/nvidia
- export LD_PRELOAD=/usr/lib/x86_64-linux-gnu/libGLEW.so