Module moog.tasks.abstract_task

Abstract task.

Classes

class AbstractTask

Expand source code

class AbstractTask(abc.ABC):
    """Abstract task class.
    
    All tasks must inherit from this class.
    """

    @abc.abstractmethod
    def reset(self, state, meta_state):
        """Reset the task.

        This should do whatever the task needs to do to itself when the
        environment resets. For example, this could reset a count of the number
        of steps taken in the episode, zero out any running state information,
        etc.
        
        Args:
            state: OrderedDict of iterables of sprites.
        """
        pass
    
    @abc.abstractmethod
    def reward(self, state, meta_state, step_count):
        """Get reward and whether to reset based on the environment state.
        
        Args:
            state: OrderedDict of iterables of sprites.
            step_count: Current number of environment steps since last reset.

        Returns:
            reward: Scalar. Reward for current state.
            should_reset: Bool. Whether environment should reset on next step.
        """
        pass

Abstract task class.

All tasks must inherit from this class.

Ancestors

• abc.ABC

Subclasses

• CompositeTask
• ContactReward
• Reset
• StayAlive

Methods

def reset(self, state, meta_state)

Expand source code

@abc.abstractmethod
def reset(self, state, meta_state):
    """Reset the task.

    This should do whatever the task needs to do to itself when the
    environment resets. For example, this could reset a count of the number
    of steps taken in the episode, zero out any running state information,
    etc.
    
    Args:
        state: OrderedDict of iterables of sprites.
    """
    pass

Reset the task.

This should do whatever the task needs to do to itself when the environment resets. For example, this could reset a count of the number of steps taken in the episode, zero out any running state information, etc.

Args

state

OrderedDict of iterables of sprites.

def reward(self, state, meta_state, step_count)

Expand source code

@abc.abstractmethod
def reward(self, state, meta_state, step_count):
    """Get reward and whether to reset based on the environment state.
    
    Args:
        state: OrderedDict of iterables of sprites.
        step_count: Current number of environment steps since last reset.

    Returns:
        reward: Scalar. Reward for current state.
        should_reset: Bool. Whether environment should reset on next step.
    """
    pass

Get reward and whether to reset based on the environment state.

Args

state

OrderedDict of iterables of sprites.

step_count

Current number of environment steps since last reset.

Returns

reward

Scalar. Reward for current state.

should_reset

Bool. Whether environment should reset on next step.