Sawyer robot learning with PPO algorithms with visual-based observations, with Unity for Sawyer robot simulation and game building.
git clone https://github.com/quantumiracle/Robot_Learning.git
cd Robot_Learning
chmod +x sawyer_visual.x86
(./sawyer_visual.x86 could see the robot env running)
conda create -n robo python=3.5
source activate robo
pip install tensorflow-gpu==1.12.0
pip install matplotlib scipy gym gym_unity
python ppo_sawyer_visual.py --train
If you meet the problem (error) of array or list with gym_unity, try:
change in conda files: /envs/robo/lib/python3.5/site-packages/gym_unity/envs/unity_env.py
line 172 in _single_step()
function add info.visual_observations=np.array(info.visual_observations)
.