pkumusic / O-DRL

Object Sensitive Deep Reinforcement Learning

O-DRL

Object Sensitive Deep Reinforcement Learning. Combining Object Features with Deep Reinforcement Learning methods.

This implementation of Deep Reinforcement Learning (DRL) methods are based on tensorflow and tensorpack. I used an older version of tensorpack and did some changes to fit into our project. For installing dependencies of tensorpack, please refer here.

Some baselines in OpenAI Gym

• DQN: Nature DQN in Human-level control through deep reinforcement learning
• DDQN: Double DQN in Deep Reinforcement Learning with Double Q-learning
• Dueling: Dueling DQN in Dueling Network Architectures for Deep Reinforcement Learning
• DDDQN: One incorporating Double and Dueling DQN

Git Configuration

Initialization

https://github.com/pkumusic/O-DRL.git

export path

export PYTHONPATH=$PYTHONPATH:path/to/O-DRL

Deploy on Mac

Dependencies of Tensorpack

• please refer tensorpack for installation. Below is what I did in my Mac.
• Python 2 or 3
• Install TensorFlow >= 0.10 pip install tensorflow Please do this before installing numpy. Since sometimes numpy would have version confliction with tensorflow version.
• Python bindings for OpenCV pip install opencv-python (This works only for CPU mode. If you're using GPU, please compile opencv with Python executable enabled. reference)
• Installing requirements for tensorpack

Additional on deploy on our turtle Machine

export LD_LIBRARY_PATH=$LD_LIBRARY_PATH:/usr/local/cuda/lib64`
export PYTHONPATH=$PYTHONPATH:`readlink -f path/to/tensorpack`
export PYTHONPATH=$PYTHONPATH:`readlink -f /usr0/home/ttian1/ml/tensorpack`