A lightweight, pure Go implementation of OpenAI's Gym for reinforcement learning environments.
go get github.com/qntx/gymRun a CartPole-v1 environment with minimal code.
env, err := gym.Make[[]float64, int]("CartPole-v1", map[string]any{
"render_mode": "human",
"sutton_barto_reward": false,
})
defer env.Close()
obs, info, err := env.Reset(context.Background(), 0, nil)
for step := 0; step < 500; step++ {
action, err := env.ActionSpace().Sample(nil, nil)
obs, reward, done, truncated, info, err := env.Step(context.Background(), action)
env.Render()
if done || truncated {
break
}
}This library provides implementations of classic reinforcement learning environments. The table below shows the current implementation status:
| Category | Environment | Rendering | Vectorized | Configurable | Action Space | Observation Space | Implemented |
|---|---|---|---|---|---|---|---|
| Classic | |||||||
CartPole-v1 |
Y | N | Y | Discrete(2) | Box(4,) | √ | |
CartPole-v0 |
Y | N | Y | Discrete(2) | Box(4,) | √ | |
Acrobot-v1 |
N | N | N | Discrete(3) | Box(6,) | ||
MountainCar-v0 |
N | N | N | Discrete(3) | Box(2,) | ||
MountainCarContinuous-v0 |
N | N | N | Box(1,) | Box(2,) | ||
Pendulum-v1 |
N | N | N | Box(1,) | Box(3,) | ||
| Box2D | |||||||
LunarLander-v2 |
N | N | N | Discrete(4) | Box(8,) | ||
LunarLanderContinuous-v2 |
N | N | N | Box(2,) | Box(8,) | ||
BipedalWalker-v3 |
N | N | N | Box(4,) | Box(24,) | ||
BipedalWalkerHardcore-v3 |
N | N | N | Box(4,) | Box(24,) | ||
CarRacing-v2 |
N | N | N | Box(3,) | Box(96,96,3) | ||
| Toy Text | |||||||
FrozenLake-v1 |
N | N | N | Discrete(4) | Discrete(16) | ||
FrozenLake8x8-v1 |
N | N | N | Discrete(4) | Discrete(64) | ||
CliffWalking-v0 |
N | N | N | Discrete(4) | Discrete(48) | ||
Taxi-v3 |
N | N | N | Discrete(6) | Discrete(500) | ||
Blackjack-v1 |
N | N | N | Discrete(2) | Tuple(32,11,2) |
Legend:
- Rendering: Supports visual rendering (human mode)
- Vectorized: Supports running multiple environments in parallel
- Configurable: Supports runtime configuration options
- Implemented: √ = Fully implemented, empty = Not implemented