def _get_goals_(self) -> D.T_agent[Space[D.T_observation]]:
# return ImplicitSpace(lambda s: s.x == self.num_cols - 1 and s.y == self.num_rows - 1)
return ImplicitSpace(
lambda s: True
if (s.x == self.num_cols - 1 and s.y == self.num_rows - 1) or
(s.x == -1 and s.y == -1) else False
) # trick to consider dead-end state as a goal to avoid modeling cycles
def _get_goals_(self) -> D.T_agent[Space[D.T_observation]]:
return ImplicitSpace(
lambda state: state.x == (self.num_cols - 1)
and state.y == (self.num_rows - 1)
)
def _get_goals_(self):
return ImplicitSpace(lambda observation: ((observation._context[
5] >= self._max_depth) or (self._termination_is_goal and (
observation._context[3].termination
if observation._context[3] is not None else False))))
def _get_goals_(self) -> Space[D.T_observation]:
return ImplicitSpace(lambda x: (x in self.targets))
def _get_action_space_(self) -> Space[D.T_event]:
return ImplicitSpace(lambda x: True)
def _get_goals_(self) -> Space[D.T_observation]:
return ImplicitSpace(lambda x: self.state_goal[x])