Skip to content

google-deepmind/dm_env_rpc

Repository files navigation

dm_env_rpc: A networking protocol for agent-environment communication.

PyPI Python version PyPI version

dm_env_rpc is a remote procedure call (RPC) protocol for communicating between machine learning agents and environments. It uses gRPC as the underlying communication framework, specifically its bidirectional streaming RPC variant.

This package also contains an implementation of dm_env, a Python interface for interacting with such environments.

Please see the documentation for more detailed information on the semantics of the protocol and how to use it. The examples sub-directory also provides examples of RL environments implemented using the dm_env_rpc protocol.

Intended audience

Games can make for interesting AI research platforms, for example as reinforcement learning (RL) environments. However, exposing a game as an RL environment can be a subtle, fraught process. We aim to provide a protocol that allows agents and environments to communicate in a standardized way, without specialized knowledge about how the other side works. Game developers can expose their games as environments with minimal domain knowledge and researchers can test their agents on a large library of different games.

This protocol also removes the need for agents and environments to run in the same process or even on the same machine, allowing agents and environments to have very different technology stacks and requirements.

Documentation

Installation

Note: You may optionally wish to create a Python Virtual Environment to prevent conflicts with your system's Python environment.

dm_env_rpc can be installed from PyPi using pip:

$ pip install dm-env-rpc

To also install the dependencies for the examples/, install with:

$ pip install dm-env-rpc[examples]

Alternatively, you can install dm_env_rpc by cloning a local copy of our GitHub repository:

$ git clone --recursive https://github.com/deepmind/dm_env_rpc.git
$ pip install ./dm_env_rpc

Citing dm_env_rpc

To cite this repository:

@misc{dm_env_rpc2019,
  author = {Tom Ward and Jay Lemmon},
  title = {dm\_env\_rpc: A networking protocol for agent-environment communication},
  url = {http://github.com/deepmind/dm_env_rpc},
  year = {2019},
}

Developer Notes

In order to run the included unit tests, developers must also install additional dependencies defined in requirements.txt, at which point the tests should be runnable via pytest. For example:

$ git clone --recursive https://github.com/deepmind/dm_env_rpc.git
$ pip install -U pip wheel
$ pip install -e ./dm_env_rpc
$ pip install -r ./dm_env_rpc/requirements.txt
$ pytest ./dm_env_rpc

Notice

This is not an officially supported Google product