Environment

tinyrl.Environment is the abstract base class for all environments. It defines the interface that every environment must implement.

Attributes

Attribute	Type	Description
state_dim	int	Dimensionality of the observation vector
n_actions	int \| None	Number of discrete actions (for discrete envs)
action_dim	int \| None	Dimensionality of the action vector (for continuous envs)
max_steps	int	Maximum steps per episode

Set n_actions for discrete action spaces, action_dim for continuous.

Abstract methods

reset() -> np.ndarray

Reset the environment to its initial state.

Returns: the initial observation as a numpy array.

obs = env.reset()

step(action: int | np.ndarray) -> tuple[np.ndarray, float, bool]

Take one step in the environment.

Args:

• action — action index (int, discrete) or action vector (np.ndarray, continuous)

Returns: a tuple of (observation, reward, done)

• observation — the new state as a numpy array
• reward — scalar reward for this transition
• done — True if the episode has ended

obs, reward, done = env.step(action)

_get_obs() -> np.ndarray

Return the current observation. Called internally by reset() and step().

render(action=None, step_num=0)

Display the current state of the environment.

Args:

• action — the action that was just taken (for display purposes)
• step_num — current step number in the episode