Just wanted to let everyone know that I just released a package from my PhD work for Deep Reinforcement Learning using TensorFlow 2.0 and ROS2 Foxy. This package takes advantage of ROS2’s DDS for both synchronous and asynchronous distributed training. It can also be used for one or more robots within an environment.
In this package, you define the networks and what constitutes a step within your environment. This package is more of a template than an actual package since things are left empty for users to actually use. This package includes popular Deep RL algorithms such as:
- Deep Q-Learning
- Double Deep Q-Learning
- Actor-Critic with Shared Network
- Actor-Critic with Dual network
I will be releasing a Proximal Policy Optimization implementation soon, so please let me know what you think and drop an issue or pull request if you would like to contribute! Credit to my wife, Amanda, for the cute robot trainer picture!