AgentPOV#

class curiosity_gym.core.agentpov.AgentPOV(action_space: Discrete, observation_space: Box, env_size: tuple[int, int])#

Abstract agent point-of-view class.

Implements the processing of inputs going into a grid environment (actions) and outputs coming out of a grid environment (observations). The transform_obs() method has to be implemented by inheriting pov classes.

Parameters:
  • action_space (gymnasium.spaces.Discrete) – The action space defining what actions can be taken by an RL agent within a grid environment.

  • observation_space (gymnasium.spaces.MultiDiscrete) – The observation space defining the structure of the observations being returned by a grid environment.

  • env_size (tuple[int, int]) – Number of (horizontal, vertical) cells in the grid environment where the pov class is used.

Methods

is_visible

Check if grid cell at given position is visible by the agent.

transform_action

Transform given action so that it is compatible with the grid environment.

transform_obs

Transform environment state to agent observation.