Gym.NET icon indicating copy to clipboard operation
Gym.NET copied to clipboard

openai/gym's popular toolkit for developing and comparing reinforcement learning algorithms port to C#.

Gym.NET

NuGet

A complete port of openai/gym to C#.
** WORK IN PROGRESS **

openai/gym

OpenAI Gym is a toolkit for developing and comparing reinforcement learning algorithms. This is the gym open-source library, which gives you access to a standardized set of environments.

Installation

### For gym's abstract classes for RL, install:
PM> Install-Package Gym.NET

### For implemented environments, install:
PM> Install-Package Gym.NET.Environments

Example

The following example runs and renders cartpole-v1 environment.

using NumSharp;
using SixLabors.ImageSharp;
using Gym.Environments;
using Gym.Environments.Envs.Classic;
using Gym.Rendering.WinForm;

CartPoleEnv cp = new CartPoleEnv(WinFormEnvViewer.Factory); //or AvaloniaEnvViewer.Factory
bool done = true;
for (int i = 0; i < 100_000; i++)
{
    if (done)
    {
        NDArray observation = cp.Reset();
        done = false;
    }
    else
    {
        var (observation, reward, _done, information) = cp.Step((i % 2)); //we switch between moving left and right
        done = _done;
        //do something with the reward and observation.
    }

    SixLabors.ImageSharp.Image img = cp.Render(); //returns the image that was rendered.
    Thread.Sleep(15); //this is to prevent it from finishing instantly !
}

Roadmap

  • Implement Spaces

    • [X] Space (base class)
    • [X] Box
    • [X] Discrete
    • [ ] multi.*.py
  • Implement Env base classes

    • [X] Env(object)
    • [ ] GoalEnv(Env)
  • Implement environments
    To run an environment, see Gym.Tests

    • [X] Convert Gym.Environments to a net-standard project.
    • [ ] classics
      • [X] CartPole-v1
        • [ ] Compare visually against python's version
      • [ ] walker2d_v3
      • [ ] acrobot
      • [ ] continuous_mountain_car
      • [ ] mountain_car
      • [ ] pendulum
      • [ ] rendering
    • [ ] Mujco
      • [ ] ant_v3
      • [ ] half_cheetah_v3
      • [ ] hopper_v3
      • [ ] humanoid_v3
      • [ ] humanoidstandup
      • [ ] inverted_double_pendulum
      • [ ] inverted_pendulum
      • [ ] mujoco_env
      • [ ] pusher
      • [ ] reacher
      • [ ] striker
      • [ ] swimmer_v3
      • [ ] thrower
    • [ ] box2d
      • [ ] bipedal_walker
      • [ ] car_dynamics
      • [ ] car_racing
      • [ ] lunar_lander
    • [ ] atari